CN115255565A - Visual sensing detection method and application of narrow gap welding groove edge based on global pattern recognition - Google Patents

Abstract

The invention discloses a visual sensing detection method and application of a narrow gap welding groove edge for global pattern recognition. ROI window, intercept the ROI window image of the groove edge, and use the image processing method to extract the position point of the groove edge on the opposite side of the arc. By looking for a subset of position points with the same abscissa on the edge of the groove, outlier data subsets that may be caused by interference such as welding spatter, arc light and smoke are eliminated; by performing linear fitting or Take the mean value and reconstruct the groove edge line containing h position points. The method of the invention realizes the precise detection of the edge position of the narrow gap welding passive vision sensing groove under the interference condition. The method of the invention is simple and convenient, the anti-interference ability is strong, the environmental adaptability is good, and the detection precision of the groove edge position is high.

Description

Vision sensing detection method of narrow gap welding groove edge based on global pattern recognition and application

technical field

The invention belongs to the field of welding technology, and in particular relates to a visual sensor detection method and application of a narrow gap welding groove edge by global pattern recognition.

Background technique

Narrow gap MIG welding is a welding method with high efficiency, high quality and low cost. In the narrow gap welding process, due to the influence of groove processing error, assembly error, welding deformation, groove form and other factors, the center of the welding torch deviates from the center of the groove, resulting in uneven penetration on both sides of the groove. Therefore, it is necessary to extract and detect the edge of the groove in real time, so as to realize the detection of the width of the groove, the detection of the center of the groove and the real-time tracking control of the welding process.

In order to achieve accurate detection and real-time tracking control of the welding process, sensing technology is the key. At present, visual sensing is widely used due to its advantages of non-contact, high sensitivity, and strong anti-interference ability. Specifically, it can be divided into active visual sensing (requiring an external light source, such as a laser sensor) and passive visual sensing (using the arc or the molten pool itself as the light source). Among them, the passive visual sensing method can achieve synchronous sensing with the arc position, no need for an external light source, low cost, suitable for various grooves, and has the advantages of obtaining a large amount of information, and is widely used. However, there are interferences such as welding spatter, arc light and welding fumes in the welding process. How to accurately detect the edge position of the groove under the interference conditions is the key to realize the accurate detection and real-time tracking control of welding process parameters.

The Chinese patent number is ZL201610517913.X, and the invention patent titled "Welding Groove Edge Position Visual Sensing Detection Method" discloses a welding groove edge position visual sensing detection method. It uses the visual sensor to collect the global image of the welding area, intercepts the edge image of the opposite side of the arc through the ROI window, and extracts the edge line of the groove after image processing; on the extracted edge line of the groove, recognizes the small window through the local pattern to detect the edge The straightest section of the edge line, and the position detection value of the groove edge on the straightest section is used as the detection value of the welding groove edge position. The disadvantages are: 1) This method belongs to the local pattern recognition method, and it is easy to fall into the local optimum. When the height of the ROI window at the edge of the groove is 100 pixels, the size of a single welding spatter can be eliminated as 1.4 mm. If the size of the welding spatter continues to increase, or When the proportion of welding spatter on the edge of the groove exceeds 60%, the detection accuracy of the edge of the groove is significantly reduced; when the height of the ROI window on the edge of the groove drops to 80pixels, the size of the welding spatter removed by this method is about 0.8mm, which shows that the accuracy of the algorithm is affected by the slope. 2) This method uses the overall threshold segmentation method to binarize the image of the ROI window at the edge of the groove. When there is uneven distribution of image gray levels caused by arc light and smoke interference, it is difficult to accurately Extract the groove edge contour line; 3) When performing adaptive positioning on the groove edge ROI window, no digital filtering algorithm is used to suppress the fluctuation of the highest point of the arc, and the accuracy of the groove edge ROI window positioning is easily affected by welding spatter and arc impact on stability.

The Chinese patent number is ZL201410741503.4, and the invention patent titled "Adaptive Control Method and Device for Narrow Gap Welding Arc Shake" discloses an adaptive control method and device for narrow gap welding arc shake. This method uses the infrared camera system to extract the groove width information in real time, and realizes the adaptive control of the arc shaking angle according to the change of the groove width; its disadvantages are: 1) This method does not use the pattern recognition method, but uses the median or mean value The filter method obtains the left and right edge positions of the groove and the detection value of the groove width. When the welding spatter accounts for more than 50%, the detection accuracy of the groove width is easily affected by the large welding spatter; 2) This method uses the overall threshold segmentation method Binarize the image of the ROI window on the edge of the groove. When there is uneven distribution of image gray levels due to arc light and smoke interference, it is difficult to accurately extract the contour line of the edge of the groove, which affects the detection accuracy of the groove width; 3) ROI When the window position is adaptively positioned, the influence of the background ratio in the ROI window image on the edge of the groove on the edge line extraction of the groove is not considered, which reduces the detection accuracy of the groove width.

The Chinese patent number is ZL201210325926.9, and the invention patent titled "Narrow Gap Welding Monitoring and Weld Deviation Detection Method Based on Infrared Vision Sensing" discloses a narrow gap welding monitoring and weld deviation detection method based on infrared vision sensing Detection method. This method uses an infrared CMOS camera to obtain welding images, and obtains the weld deviation value by calculating the distance from the left (or right) edge of the molten pool away from the arc side to the left (or right) boundary of the left (or right) interception window. The disadvantages are: 1) This method does not adopt the adaptive positioning method when setting the interception windows of the left and right groove edge positions, and its adaptability is poor; 2) This method does not solve the interference of welding smoke and spatter on the groove edge position detection. impact on precision.

Contents of the invention

The purpose of the present invention is to address the shortcomings of the existing technology such as low detection accuracy of the edge position of the narrow gap welding groove, poor adaptability to the welding environment, and insufficient self-adaptive ability, and proposes a high detection accuracy, strong anti-welding interference ability, A narrow gap welding groove edge visual sensing detection method with good practicability and global pattern recognition and its application.

The method of the present invention intercepts the groove edge ROI window image through the position-adaptive groove edge ROI window, extracts the groove edge original position point on the opposite side of the arc through local image processing, and searches for the groove edge original position point Groove edge position points on the vertical line segment, outlier position points on the non-perpendicular line segment are eliminated, and the remaining real position points are reconstructed for the groove edge to achieve accurate detection of the groove edge position under welding interference conditions .

In order to achieve the above object, the present invention adopts the following technical solutions to achieve.

A visual sensor detection method for narrow-gap welding groove edges based on global pattern recognition, based on a detection system that mainly includes: an infrared camera 8, a filter system 7, a signal trigger 9, an image transmission data line 10, and a computer image processing system 11 , wherein the signal trigger 9 is connected to the infrared camera 8, the filter system 7 is coaxially installed on the lens of the infrared camera 8, and the computer image processing system 11 communicates with the infrared camera 8 through the image transmission data line 10 The infrared camera 8 is connected; the overall image 13 of the narrow gap welding area is collected by the infrared camera 8, the filter system 7 and the signal trigger 9; it is characterized in that the detection method specifically includes the following steps:

1) Determine the initial lateral position of the origin of the ROI window on the edge of the groove: the computer image processing system 11 extracts the position of the highest point 14 of the arc 2 from the global image 13 through global image processing, and determines the SROI that can intercept the edge of the groove on the same side of the arc The position of the window 16, or the position of the BROI window 17 that can capture both edges of the groove at the same time; by performing preset window image processing on the image captured by the SROI window 16, to obtain the initial groove edge position on the same side of the arc, or by analyzing the BROI The image intercepted by the window 17 is processed by the preset window image, and the initial groove edge positions on the same side and the opposite side of the arc are obtained, so as to adaptively determine the initial lateral position of the origin 15 of the ROI window 12 on the opposite side of the arc;

根据电弧最高点位置和坡口边缘ROI窗口12原点 15的初始横向位置自适应确定坡口边缘ROI窗口12的位置后，通过该自适应定位方法确定的坡口边缘ROI窗口12，截取电弧2对侧的坡口左边缘4或坡口右边缘5的ROI窗口图像，并通过局部图像处理提取坡口左边缘4或坡口右边缘5，获取由h个数据组成的坡口边缘原始位置点集合

其中，i表示当前操作，h表示ROI窗口12的高度值；2) Obtain the set of original position points on the groove edge

After the position of the groove edge ROI window 12 is adaptively determined according to the position of the highest point of the arc and the initial lateral position of the origin 15 of the groove edge ROI window 12, the arc 2 pair is intercepted through the groove edge ROI window 12 determined by the adaptive positioning method The ROI window image of the left edge 4 of the groove or the right edge 5 of the groove on the side, and extract the left edge 4 or the right edge 5 of the groove through local image processing, and obtain the original position point set of the groove edge composed of h data

Wherein, i represents the current operation, and h represents the height value of the ROI window 12;

中的位置点进行一次数据滤波：沿坡口边缘ROI 窗口12的高度h方向上，依次计算坡口边缘原始位置点集合

中相邻两位置点的x坐标的差值，将差值为零的位置点分到同一个位置点子集合中；相应地，p(1≤p<h)个位置点子集合

被找到，从而形成包含p个位置点子集合的坡口边缘预处理位置点集合

相应地，形成p个垂线段，每一个垂线段含有相同的x坐标值，即横坐标值；3) Set the original position points of the obtained groove edge

Perform a data filtering on the position points in the groove edge: along the direction of the height h of the ROI window 12 on the edge of the groove, calculate the original position point set of the edge of the groove in sequence

The difference between the x-coordinates of two adjacent position points in , and the position points with a difference of zero are divided into the same position point subset; correspondingly, p(1≤p<h) position point subsets

is found, thus forming a groove edge preprocessing position point set containing p position point subsets

Correspondingly, p vertical line segments are formed, and each vertical line segment contains the same x-coordinate value, that is, the abscissa value;

中的位置点进行二次数据滤波：具体包括如下步骤：4) Preprocessing the position point set on the edge of the groove

Perform secondary data filtering on the position points in : specifically include the following steps:

中数据的分散度：针对步骤3)获取的坡口边缘预处理位置点集合

首先判断坡口边缘预处理位置点集合

中位置点的分散度；具体为：搜索坡口边缘预处理位置点集合

中所有位置点横坐标值

的最大值x_{k_max}和最小值x_{k_min}，计算极差R_k＝x_{k_max}-x_{k_min}；判断极差R_k与数据分散度阈值R_t的大小，数据分散度阈值R_t＝INT(h×4％)，若满足R_k≤R_t，将坡口边缘预处理位置点集合

作为最终坡口边缘真实位置点集合

若R_k＞R_t时，将坡口边缘原始位置点集合

中h个位置点横坐标值

的中值

并以

作为滤波阈值变量M_k，再结合方向滤波器，滤除离群数据后，获取坡口边缘保留位置点集合

① Judging the set of preprocessing position points on the edge of the groove

Dispersion of the data in: the set of preprocessing position points for the groove edge obtained in step 3)

Firstly, determine the set of preprocessing position points on the edge of the groove

Dispersion of mid-position points; specifically: search for the set of pre-processing position points on the edge of the groove

The abscissa values of all position points in

The maximum value x _{k_max} and the minimum value x _{k_min} , calculate the range R _k ＝x _{k_max} -x _{k_min} ; judge the size of the range R _k and the data dispersion threshold R _t , the data dispersion threshold R _t = INT(h×4 %), if satisfying R _k ≤ R _t , set the groove edge preprocessing position points

As the set of real position points of the final groove edge

If R _k > R _t , collect the original position points of the groove edge

The abscissa values of the h position points

median of

and

As the filtering threshold variable M _k , combined with the direction filter, after filtering outlier data, obtain the set of reserved position points on the edge of the groove

判断位置点分散度：通过计算坡口边缘保留位置点集合

中每个位置点横坐标值

的极差R_k，如果满足R_k≤R_t，坡口边缘保留位置点集合

作为坡口边缘真实位置点集合

否则，计算坡口边缘保留位置点集合

中位置点横坐标值

的中值

并以

作为滤波阈值变量M_k。通过方向滤波器，获取过滤后的坡口边缘保留位置点集合

重复该过程，直至

中每个位置点横坐标值极差 R_k满足R_k≤R_t，形成坡口边缘真实位置点集合

②Reserve the set of position points for the edge of the groove obtained in step ①

Judging the degree of dispersion of position points: by calculating the edge of the groove to retain the set of position points

The abscissa value of each position point in

The range R _k of , if R _k ≤ R _t is satisfied, the groove edge retains a set of position points

As a set of real position points on the edge of the groove

Otherwise, calculate the groove edge reserved position point set

The abscissa value of the middle position point

median of

and

As the filtering threshold variable M _k . Through the direction filter, obtain the filtered groove edge reserved position point set

Repeat this process until

The extreme difference R _k of the abscissa value of each position point in , satisfies R _k ≤ R _t , forming a set of real position points on the edge of the groove

统计保留的真实位置点的个数N_s；当N_s≤INT(h×10％)时，采用均值计算的方法重构包含h个坡口边缘位置点的坡口边缘重构位置点集合

否则，通过最小二乘线性拟合方法，重构包含h个坡口边缘位置点的坡口边缘重构位置点集合

其中，坡口边缘位置点的横坐标为

5) Reconstructing the groove edge line: for the set of real position points of the groove edge obtained in step 4)

The number N _s of the real position points retained by statistics; when N _s ≤ INT(h×10%), the method of mean value calculation is used to reconstruct the groove edge reconstruction position point set containing h groove edge position points

Otherwise, by the least squares linear fitting method, reconstruct the groove edge reconstruction position point set containing h groove edge position points

Among them, the abscissa of the groove edge position point is

通过计算h个坡口边缘位置点横坐标的均值或拟合值，获取坡口边缘位置采样值；针对坡口边缘位置采样值采用数字滤波方法，获取坡口边缘位置检测值；6) Obtain the detection value of groove edge position: for the groove edge reconstruction position point set containing h groove edge position points obtained in step 5)

Obtaining the sampling value of the groove edge position by calculating the mean value or fitting value of the abscissa of the h groove edge position points; adopting a digital filtering method for the groove edge position sampling value to obtain the groove edge position detection value;

7) Repeat the above steps 2) to 6) until the welding process ends.

Further preferably, in the step 1), the global image processing includes: first adopting a median filter to the global image 13 to perform image denoising; After the pixel frequency is counted, global thresholding is performed to binarize the global image 13; finally, the arc contour is extracted through morphological operations, and the coordinate value of the highest point of the arc is obtained through grayscale search. The preset window image processing includes performing median filter denoising on the intercepted window image, contrast stretching to improve image contrast, using Otsu threshold processing on the window image, and then removing isolated points through morphological operations, and then using Canny edge operator Extract the groove edge line in the preset window image;

Further preferably, in said step 1), the longitudinal positions of the SROI window 16 and the BROI window 17 are determined by the ordinate value of the highest point 14 of the arc; and the set distance from the abscissa value of the arc highest point 14 to the groove edge.

Further preferably, in the step 2), the adaptive positioning method of the groove edge ROI window 12 includes:

① From the global image 13 of the latest N ₁ consecutive frames, extract the ordinate values of N ₁ arc highest points 14, and then obtain the filter value F ₁ of the N ₁ ordinate values by digital filtering method, and use (F _{1 -} δ1) as the ordinate value of the origin 15 of the groove edge ROI window 12; wherein, δ1 is _a correction constant of the ordinate position value of the groove edge ROI window 12 origin 15, and its value range is [-h,h];

② Extract N ₂ position values of the left edge 4 of the groove or the right edge 5 of the groove from the recent N ₂ previous frame groove edge ROI window 12 images on the same side as the current groove edge to be detected, and then pass the digital The filter method obtains the filter value F ₂ of its N ₂ position values, and uses (F ₂ -δ ₂ ) as the abscissa value of the origin 15 of the groove edge ROI window 12; or for the current global image 13, through the The BROI window 17, which can simultaneously intercept the double edges of the groove, extracts the current position values of the left edge 4 of the groove and the right edge 5 of the groove, and then obtains the filter value F of the groove edge position values of the nearest N ₃ opposite sides of the arc through a digital filtering method ₃ , and take (F ₃ -δ ₂ ) as the abscissa value of the origin 15 of the groove edge ROI window 12; wherein, δ ₂ is a correction constant for the abscissa position value of the groove edge ROI window 12 origin 15 , is the half-width of the groove edge ROI window 12 , that is, δ ₂ =w/2, where w is the width of the groove edge ROI window 12 .

Further preferably, in the step 2): the local image processing includes performing median filtering and denoising on the groove edge ROI window image, then performing contrast stretching to improve the contrast of the groove edge ROI window image, and then performing partition threshold , obtain the binarized image, use morphological operations to further denoise, and extract the edge line of the groove edge ROI window image.

Wherein the contrast stretching selects the piecewise linear change function, the morphological operation selects the closed operation, the edge extraction selects the Canny operator, and the partition threshold is after the groove edge ROI window image is divided into 4 equal parts on average, Otsu method thresholding is applied to the image of each partition separately.

或坡口边缘保留位置点集合

中的位置点横坐标值。Further preferably, in the step 4): the direction filter is expressed as: when the groove edge position point is located on the left edge, the direction filter is a low-pass filter, expressed as x _{k ≤} M _k , That is, retain the position point whose abscissa value is less than or equal to the filter threshold variable M _k ; when the groove edge position point is located on the right edge, the direction filter is a high-pass filter, expressed as x _k ≥ M _k , that is, retain The abscissa value of the position point is greater than or equal to the position point of the filter threshold variable M _k ; x _k represents the circular data variable, and its value is equal to the preprocessing position point set of the edge of the groove

or bevel edge retaining location point set

The abscissa value of the location point in .

Further preferably, in the step 6), the digital filtering method is limiting and anti-pulse average filtering, which specifically includes the following steps:

① Determine the number of data _{Nf included in the filtering window: the filtering window includes the latest N f sampling} values of the edge position of the groove including the sampling value G _s of the edge position of the groove this time, and the filtering window includes the number of data N _f >2;

② Calculate the current sampling deviation η ₁ : Calculate the average value E _s of the sampling values of the N _f groove edge positions within the current filtering window, and calculate the absolute value of the difference between the current groove edge position sampling values G _s and the average value E _s , as this sampling deviation η ₁ ;

③Calculate this sampling deviation degree d _e : calculate the ratio of this sampling deviation η ₁ to the previous sampling deviation η ₀ , and use this ratio as this sampling deviation degree d _e ;

④ Repair abnormal sampling value: when the sampling deviation d _e is greater than the abnormal repair threshold d _th , the previous detection value of the edge position of the groove is used as the sampling value G _s of the edge position of the groove this time to realize the abnormal sampling of this time Value repair, the abnormal repair threshold d _th =3～10;

⑤Calculate the detection value G _d of the groove edge position this time: after removing the maximum value and the minimum value from the N _f groove edge position sampling values, calculate the mean value for the remaining (N _f -2) groove edge position sampling values , and use this mean value as the groove edge position detection value G _d this time.

A kind of global pattern recognition of the present invention narrow gap welding groove edge visual sensing detection method is applied to the detection of welding groove width, groove center position and real-time tracking control of welding process; the applied method is, based on the detected For the left edge 4 of the welding groove and the right edge 5 of the groove, the center position value of the welding groove is obtained by calculating the mean value of the corresponding positions of the two edges; the width value of the welding groove is obtained by calculating the difference between the corresponding positions of the two edges.

Compared with prior art, advantage and beneficial effect of the present invention mainly are:

1) Through the visual sensor detection method of the narrow gap welding groove edge by global pattern recognition, the outlier data is filtered on the original position point of the groove edge, which can remove the simulated welding spatter below about 2.4mm, and improve the detection accuracy of the groove edge and the adaptability of the proposed method to the welding environment;

2) Through the adaptively positioned groove edge ROI window, according to the ordinate value of the arc highest point and the digital filtering result of the groove edge position value in the previous frame on the same side, adaptively obtain the ordinate value and the origin position of the groove edge ROI window The abscissa value reduces the impact of welding spatter interference and arc stability on the positioning accuracy of the ROI window at the edge of the groove; through welding tests, it is shown that the maximum position difference of the arc highest point between two adjacent frames of welding images is reduced from ~4mm before filtering, Stable to within 1mm; improves the effectiveness of ROI window positioning.

3) Through the self-correction method of the groove edge image position, the background ratio in the groove edge ROI window image is kept at ~50%, which improves the separation of the groove edge image and the extraction accuracy of the groove edge;

4) Through the quarter-part partition threshold method of the ROI window image on the edge of the groove, under the influence of welding fume and arc light, the dispersion of the edge position points of the groove is reduced, and the extraction accuracy of the edge of the groove is increased by ~68%.

Description of drawings

Figure 1 is a block diagram of a passive visual sensing detection system for narrow gap groove edges;

Fig. 2 is a schematic diagram of the global image of the welding area collected by the detection system shown in Fig. 1; Fig. 2(a) is a schematic diagram of the global image with the arc at the right edge of the groove; Fig. 2(b) is a global image with the arc at the left edge of the groove Image schematic.

Figure 3 is a schematic diagram of the SROI and BROI windows in the global image of the narrow gap welding area;

Fig. 4 is a flow chart of narrow gap welding image processing;

Fig. 5 is a schematic diagram of the global pattern recognition algorithm of the groove edge for passive visual sensor detection of narrow gap welding; Fig. 5(a) is the primary data filtering, and Fig. 5(b) is the secondary data filtering;

Fig. 6 is an example of image processing for narrow gap welding during pulse welding. Figure 6(a) and (b) are the global images of the welding area, Figure 6(c) and (d) are the images of the extracted arc area, Figure 6(e) and (f) are the right and left edge ROI windows of the bevel, Fig. 6 (g) and (h) are the ROI window images and image processing results of the right and left edges of the groove, and Figure 6 (j) and (j) are the extracted right and left edges of the groove;

Fig. 7 is an embodiment of the global pattern recognition algorithm for the distribution of raw data on the right edge of the groove during pulse welding. 7(a) The original position point of the groove edge, 7(b) Search the outlier position point in the original position point of the groove edge, 7(c) The preprocessing position point of the groove edge, 7(d) The real position of the groove edge Point, 7(e) reconstructed groove edge line;

Fig. 8 is an embodiment of narrow gap welding image processing during DC welding; Fig. 8 (a) and (b) are the global images of the welding area, Fig. 8 (c) and (d) are the extracted arc area images, and Fig. 8 (e) and (f) are the right and left edge ROI windows of the groove, Figure 8(g) and (h) are the images and image processing results of the right and left edge ROI windows of the groove, and Figure 8(j) and (j) are the extracted Groove right and left edges;

Fig. 9 is the embodiment of the global pattern recognition algorithm for the original data distribution of the right edge of the groove during DC welding; 9 (a) the original position point of the groove edge, and 9 (b) search for the outlier position point in the original position point of the groove edge , 9(c) primary reserved position point of groove edge, 9(e) secondary reserved position point of groove edge, 9(e) real position point of groove edge, 9(f) reconstructed groove edge line;

为本次的坡口边缘ROI窗口12的原点；h—ROI窗口12 的高度；

和

—坡口左、右边缘位置点。In Figure 1, Figure 2 and Figure 3: 1a—left wall of welding groove; 1b—right wall of welding groove; 2—electric arc; 3a—wire feeder; 3b—arc motion driver; 3c—conductive rod mechanism; 3d—welding wire; 4—left edge of groove; 5—right edge of groove; 6—welding groove; 7—filter system; 8—infrared camera; 9—signal trigger; 10—image transmission data line; 11— Computer image processing system; 12—ROI window on groove edge; 13—global image of welding area; 14—highest point of arc; 15—origin of ROI window 12 on groove edge; 16—SROI window; 17—BROI window; V _w — Welding speed; θ—shooting angle of infrared camera 8; P _arc —shaking arc position signal; i _b —base value current signal; O _i-1 —origin of the global image of the welding area in the previous frame, which is also the rectangular coordinate system x _i- The origin of ₁ -O _i-1 -y _i-1 ; O _i —the origin of the global image of the welding area in the current frame is also the origin of the Cartesian coordinate system x _i -O _i -y _i ;

Be the origin of the groove edge ROI window 12 this time; h—the height of the ROI window 12;

and

—The position points of the left and right edges of the groove.

Figure 2(a) in Figure 2 corresponds to the previous (i-1th) sampling, indicating the situation when the arc is on the right; Figure 2(b) corresponds to the current (i-th) sampling, indicating that the arc is at situation on the left. In the figure, the black back-shaped closed solid line represents the outline of the welding image, including the left and right side walls of the groove and the front and rear edges of the molten pool.

Detailed ways

Below in conjunction with accompanying drawing and specific embodiment, the technical scheme of the present invention is described in further detail, but the scope of protection of the present invention is not limited to the following examples, all technical schemes obtained in the form of equivalent replacement or equivalent transformation are protected by the present invention within range.

As shown in Figure 1, it is a block diagram of a passive visual sensing detection system for narrow gap groove edges used in the narrow gap welding groove edge visual sensing detection method for global pattern recognition of the present invention, which mainly includes: narrow gap shaking/rotation Arc welding torch, vision sensing system and computer image processing system 11. Among them, the narrow gap shaking/rotating arc welding torch includes an arc motion driver 3b and a conductive rod mechanism 3c, the arc motion driver 3b includes a driving mechanism and a welding torch controller, and its conductive rod mechanism 3c is composed of a bent conductive rod and a corresponding The bending conductive rod mechanism composed of a straight conductive tip connected to it, or the eccentric conductive rod mechanism composed of a straight conductive rod and an eccentric conductive tip connected to it; the welding wire 3d passes through the wire feeder 3a and passes through the arc motion driver After 3b, it protrudes obliquely from the conductive rod mechanism 3c, generates a welding arc 2 in the welding groove 6, and performs periodic circular arc shaking or unidirectional rotation around the center line of the welding torch. When the arc shakes or rotates to the nearest position on the left side wall of the groove or the right side wall of the groove, the torch controller sends out the arc position signal P _arc ; When staying at the nearest position of the wall, the welding torch controller detects the arc current in real time through the current sensor, and generates the arc base value current signal i _b .

The visual sensing system includes a filter system 7 , an infrared camera 8 , a signal trigger 9 , and an image transmission data line 10 . Wherein, the filter system 7 includes a central wavelength range of 800 to 1100nm narrow-band filter, a neutral light-reducing mirror, and a UV mirror; ~60°) aim at the welding pool, keep a fixed distance from the welding torch, and move forward at the welding speed V _w , the focal length range of the infrared camera 8 is 18-45mm, and the aperture range is f/5.6-36; the signal trigger 9 After receiving the arc position signal P _arc or the arc base value current signal i _b at the same time, trigger the infrared camera 8 through an external trigger method, start to collect the global image 13 of the welding area, and transmit data through the image connected to the infrared camera 8 The wire 10 is sent into the computer image processing system 11 for subsequent detection of the left edge 4 of the welding groove and the right edge 5 of the groove.

The general scheme of the narrow gap welding groove edge visual sensor detection method for global pattern recognition of the present invention is: the computer image processing system 11 is not triggered to obtain the global image 13 of the welding area, and the position of the highest arc point 14 is obtained by global image processing, As shown in Figure 2; through the groove edge ROI window 12 adapted to the arc position and the groove edge position, the groove edge ROI window image on the opposite side of the arc 2 is intercepted, and after the groove edge ROI window image is processed, the extracted The left edge of the groove or the right edge line of the groove form the original position point of the groove edge corresponding to the height of the ROI window of the groove edge; for the obtained original position point of the groove edge, based on the overall pattern recognition, an outlier Data filtering algorithm, by calculating the difference of the abscissa of adjacent position points and the direction filter, eliminating the outlier position points disturbed by welding, and reconstructing the edge of the groove by linear fitting or calculating the mean value of the retained real position points Line, to achieve accurate detection of the edge position of the groove under the condition of welding interference. Its specific steps include:

Step 1): Determine the adaptive positioning of the groove edge ROI window, see Fig. 2, Fig. 3 and Fig. 4, specifically include the following steps:

①The computer image processing system 11 collects the global image 13, as shown in Figure 2; sequentially through median filtering, histogram analysis, global threshold and morphological operations, the global image 13 is processed globally, as shown in Figure 4; Extract the arc 2 area image from the global image 13, perform a grayscale search on the acquired arc 2 area image, and find out the abscissa value and ordinate value of the highest point 14 of the arc;

② Through the ordinate value of the highest point 14 of the arc, and the distance from the abscissa value of the highest point 14 of the arc to the groove edge on the same side of the arc, determine the position of the SROI window 16 that can intercept the groove edge on the same side of the arc in the global image 13, Or also according to the width of the global image 13, determine the position of the BROI window 17 that can intercept the double edge of the groove simultaneously, as shown in Figure 3;

③ For the preset window images intercepted by SROI window 16 and BROI window 17, the median filter, contrast stretching, global threshold, morphological operation and Canny edge extraction of the window images are respectively performed in sequence, as shown in Figure 4, to obtain the arc 2 The initial groove edge position on the same side, or obtain the groove edge positions on the same side and the opposite side of the arc, so as to adaptively determine the initial lateral position of the origin 15 of the ROI window 12 on the opposite side of the arc.

Wherein the specific method steps of determining the initial lateral position of the origin 15 of the ROI window 12 of the groove edge on the opposite side of the arc adaptively are:

① By extracting the ordinate values of N ₁ arc highest points 14 from the global image 13 of the latest continuous N ₁ frames, and then obtaining the filter value F ₁ of the N ₁ ordinate values by digital filtering method, and using ( F ₁ -δ ₁ ) as the ordinate value of the origin 15 of the groove edge ROI window 12; wherein, δ ₁ is a correction constant for the ordinate position value of the groove edge ROI window 12 origin 15, preferably δ ₁ is Half of the height of the ROI window 12 is the half height of the groove edge ROI window 12, at this time δ ₁ =h/2, where h is the height of the groove edge ROI window 12;

② Extract N ₂ position values of the left edge 4 of the groove or the right edge 5 of the groove from the recent N ₂ previous frame groove edge ROI window 12 images on the same side as the current groove edge to be detected, and then pass the digital The filter method obtains the filter value F ₂ of its N ₂ position values, and uses (F ₂ -δ ₂ ) as the abscissa value of the origin 15 of the groove edge ROI window 12; or for the current global image 13, through the The BROI window 17, which can simultaneously intercept the double edges of the groove, extracts the current position values of the left edge 4 of the groove and the right edge 5 of the groove, and then obtains the filter value F of the groove edge position values of the nearest N ₃ opposite sides of the arc through a digital filtering method ₃ , and take (F ₃ -δ ₂ ) as the abscissa value of the origin 15 of the groove edge ROI window 12; wherein, δ ₂ is a correction constant for the abscissa position value of the groove edge ROI window 12 origin 15 , preferably half of the width of the groove edge ROI window 12, that is, the half width of the groove edge ROI window 12, at this time δ ₂ =w/2, w is the width of the groove edge ROI window 12; The height of the groove edge ROI window 12 is h=80 pixels, and the width w=80 pixels.

其中，所述对比度拉伸优选采用分段线性变化函数、所述等份分区阈值优选采用4等份分区的大津法阈值算法、所述形态学运算优选闭运算、所述边缘提取优选采用Canny算子。Step 2): Obtain the original position point set of the groove edge, and its distribution is shown in Figure 5; after the position of the groove edge ROI window 12 is determined adaptively according to the position of the highest point of the arc and the position of the groove edge on the same side, pass through the groove edge ROI The window 12 intercepts the groove edge ROI window image of the groove left edge 4 or the groove right edge 5 on the opposite side of the arc 2, and performs local image processing on the groove edge ROI window image, including median filtering, contrast stretching, Equal partition threshold, morphological operation and edge extraction, as shown in Figure 4, on the height h direction of the ROI window 12 of the groove edge, extract the groove left edge 4 or the groove right edge 5, and obtain h A collection of original position points on the edge of the groove composed of data

Wherein, the contrast stretching preferably adopts a piecewise linear change function, the equal partition threshold preferably adopts the Otsu method threshold algorithm of four equal partitions, the morphological operation preferably uses a closed operation, and the edge extraction preferably adopts a Canny algorithm. son.

依次对两相邻的位置点x坐标值做求差运算，并将其差值为零的两个及以上相邻位置点归入同一个垂直分布子集合，形成包含p(1≤p<h)个垂直分布子集合的坡口边缘预处理位置点集合

并从坡口边缘原始位置点集合

中搜索与p个子集合中横坐标相同的坡口边缘位置点原位保留；最后将剩余的不属于任何一个子集合的位置点滤除。其中，子集合

和

中的任一边缘位置点分别表示为

和

δ、ξ和ψ为坡口边缘位置点在坡口边缘原始位置点集合

中的y向位置序号。Step 3): a data filtering, see Fig. 5; in the height h direction of the groove edge ROI window 12, for the obtained groove edge original position point set

Perform a difference operation on the x-coordinate values of two adjacent position points in turn, and classify two or more adjacent position points whose difference is zero into the same vertical distribution subset to form a set containing p(1≤p<h ) Groove edge preprocessing position point set of vertically distributed subsets

And from the original position point collection on the edge of the groove

Search for the groove edge position points with the same abscissa coordinates in the p subsets and keep them in situ; finally filter out the remaining position points that do not belong to any subset. Among them, the sub-collection

and

Any edge position point in is expressed as

and

δ, ξ and ψ are the set of groove edge position points at the original position points of the groove edge

The y-direction position number in .

表示循环过滤后剩余的子集合数。针对上述一次数据过滤获取的坡口边缘预处理位置点集合

相应地，通过寻找坡口边缘预处理位置点集合

或坡口边缘保留位置点集合

中的横坐标值x_k的最大值x_{k_max}和最小值x_{k_min}，计算极差R_k＝(x_{k_max}- x_{k_min})。当R_k≤R_t时，表明集合中位置点的分散度较小，则将坡口边缘预处理位置点集合

或坡口边缘保留位置点集合

直接作为坡口边缘真实位置点集合

否则，通过对坡口边缘预处理位置点集合

或坡口边缘保留位置点集合

进行重复迭代过滤处理，直至满足条件 R_k≤R_t时，将坡口边缘保留位置点集合

作为坡口边缘真实位置点集合

最后，对坡口边缘真实位置点集合

中的数据进行线性拟合或计算均值的方法，形成包含h个拟合数据或均值数据的坡口边缘重构位置点集合

试验中，数据分散度阈值R_t取值为INT(h×4％)。Step 4): Secondary data filtering, see Figure 5; after k times of cyclic filtering, the set of retained position points on the edge of the groove is obtained as

Indicates the number of subsets remaining after loop filtering. A collection of groove edge preprocessing position points obtained for the above-mentioned data filtering

Correspondingly, by finding the groove edge preprocessing position point set

or bevel edge retaining location point set

The maximum value x _{k_max} and the minimum value x _{k_min} of the abscissa value x _k in the calculation range R _k =(x _{k_max} - x _{k_min} ). When R _k ≤ R _t , it indicates that the dispersion degree of the position points in the set is small, then the groove edge preprocesses the position point set

or bevel edge retaining location point set

directly as the set of real position points on the groove edge

Otherwise, by preprocessing the location point set on the groove edge

or bevel edge retaining location point set

Perform repeated iterative filtering until the condition R _k ≤ R _t is satisfied, and the edge of the groove is retained as a set of position points

As a set of real position points on the edge of the groove

Finally, the set of real position points on the groove edge

The method of performing linear fitting or calculating the mean value of the data in , forming a set of reconstructed position points on the edge of the groove containing h fitting data or mean value data

In the experiment, the value of data dispersion threshold R _t is INT(h×4%).

中任一位置点的横坐标

作为数据集合X_k中的任一循环数据变量x_k，其中v(1≤v≤h)表示坡口边缘预处理位置点集合

中的数据在坡口边缘原始位置点集合

中的y向位置序号；计算坡口边缘原始位置点集合

中h个位置点横坐标中值

并以

作为滤波阈变量M_k。当k>1时，以坡口边缘保留位置点集合

中任一数据的横坐标

作为循环数据变量x_k，其中ω(1≤ω≤h)表示坡口边缘保留位置点集合

中的数据在坡口边缘原始位置点集合

中的y向位置序号；计算坡口边缘保留位置点集合

的f个子集合中的所有位置点横坐标中值

并以

作为滤波阈变量M_k。Specifically, when the number of cyclic filtering k=1, the set of position points is preprocessed with the edge of the groove

The abscissa of any point in

As any cyclic data variable x _k in the data set X _k , where v(1≤v≤h) represents the set of groove edge preprocessing position points

The data in the set of original position points on the groove edge

The y-direction position number in ; calculate the set of original position points on the edge of the groove

The median value of the abscissa of the middle h position points

and

As the filtering threshold variable M _k . When k>1, the set of position points is reserved by the edge of the groove

The abscissa of any data in

As a cyclic data variable x _k , where ω(1≤ω≤h) represents the set of reserved position points on the edge of the groove

The data in the set of original position points on the groove edge

The y-direction position number in ; calculate the groove edge reserve position point set

The abscissa median value of all position points in the f subsets

and

As the filtering threshold variable M _k .

统计保留的真实位置点的个数N_s；当N_s≤INT(h×10％)时，采用均值计算的方法重构包含h个位置点的坡口边缘重构位置点集合

否则，通过最小二乘线性拟合方法，将拟合后的h个位置点作为坡口边缘重构位置点集合

Step 5) Reconstructing the groove edge line: for the set of real position points of the groove edge obtained in the above step 4)

Statistically retain the number N _s of real position points; when N _s ≤ INT(h×10%), use the mean value calculation method to reconstruct the groove edge reconstruction position point set containing h position points

Otherwise, through the least squares linear fitting method, use the fitted h position points as the groove edge reconstruction position point set

Step 6) Calculate the detection value of the edge position of the groove: obtain the sampling value of the edge position of the groove through the mean value or fitting value of the reconstructed h position points; use the digital filtering method to obtain the sampling value of the edge position of the groove Edge position detection value.

Four specific embodiments of the narrow-gap welding groove edge visual sensing detection method for global pattern recognition of the present invention are provided below.

Example 1

As shown in Fig. 6, it is an embodiment of narrow gap welding image processing during pulse welding. The test conditions include: using a CMOS infrared camera as the infrared camera 8, after the signal trigger 9 receives the arc position signal _Parc and the arc base value current signal i _b , triggers the infrared camera 8 through an external trigger, and starts to collect the global image of the welding area 13 , and sent to the computer image processing system 11 through the image transmission data line 10 connected with the infrared camera 8, for subsequent detection of the groove width and groove center. The shooting angle θ=25° of the infrared camera 8, the aperture is 16, and the exposure time is 0.3ms; the filter system 7 includes a UV mirror, a neutral light-reducing mirror with a transmittance of 30%, and a center wavelength of 970nm, Narrowband filters with a bandwidth of 20nm. The base metal of the welded specimen is Q370qE steel plate, the gap of welding groove 6 is 11.75mm; the average current of pulse arc 2 is 322.5A, the average arc voltage is 29.3V, the welding speed V _w is 204mm/min, and the dry elongation of welding wire is 20mm, the diameter of welding wire 3d is 1.2mm, and the flow rate of welding shielding gas Ar-20% _CO2 is 25L/min. The arc-shaped shaking frequency of arc 2 is 2.5Hz, the shaking angle is 64°, and the residence time of both sides of the groove is 100ms.

Fig. 6 (a) and (b) are the global images 13 of two adjacent frames of the welding area when the electric arc 2 collected by the computer image processing system 11 shakes to the left edge 4 of the groove and the right edge 5 of the groove respectively, and the image size is 544×544 pixels. Among them, Fig. 6(a) is the global image 13 of the welding area collected when the arc 2 stays on the left edge 4 of the groove. There is a weld spatter. Fig. 6(b) is a global image 13 of the welding area collected when the arc 2 stays on the right edge 5 of the groove. After the computer image processing system 11 receives the global image 13 of the welding area, by analyzing the gray histogram of the global image 13, the gray histogram of the global image 13 of the welding area is established, and the gray value of 250 is selected as the global threshold for welding The regional global image 13 (Fig. 6(a) and 6(b)) is binarized, and the extracted arc 2 region images are shown in Fig. 6(c) and 6(d), respectively.

By performing a grayscale search on the arc 2 region images in Figures 6(c) and 6(d), it is calculated that the highest point 14 of the arc 2 in the adjacent two frame images is in the Cartesian coordinate system x _i-1 -O _i- The horizontal and vertical coordinates of ₁ -y _i-1 and x _i -O _i -y _i are (167, 390) and (371, 393) respectively. At this time, the ordinate value of the origin 15 of the groove edge ROI window 12 is obtained by moving the ordinate value of the arc 2 up by δ ₁ (δ ₁ =h/2), and the abscissa value is obtained by moving the value of the ordinate of the arc 2 up through the groove edge of the previous frame on the same side. The position is obtained by subtracting δ ₂ (δ ₂ =w/2); therefore, the coordinate values of the origin 15 of the ROI window 12 on the right and left edges of the groove on the opposite side of the arc are (385, 350) and (61, 353) respectively. The height h and width w of the ROI window are both 80 pixels, as shown by the small white boxes in Figure 6(e) and 6(f), respectively. In fact, in the detection of continuous welding images, the origin coordinate value of the ROI window on the edge of the groove is determined by the filter value of the ordinate of the highest point _of the continuous N1 arc and the filter value of _N2 consecutive edge positions of the groove on the same side , which can reduce the impact of the sudden change in the position of the ROI window on the groove edge in two adjacent frames of images on the groove edge positioning and groove width detection accuracy.

Figures 6(g) and 6(h) show the ROI window images of the right and left edges of the groove after median filtering, contrast stretching, and quadrangular partition thresholding. The background in the above-mentioned groove edge ROI window images than about 50%. Figures 6(i) and 6(j) show the extraction of the right edge 5 of the groove and the left edge 4 of the groove after the morphological operation and the Canny edge operator, as shown by the white line in the figure. It can be seen from Fig. 6(i) that there is welding spatter on the right edge 5 of the groove extracted in the ROI window image 12, so an effective method must be adopted in the subsequent detection of the groove edge in order to effectively avoid the impact of welding spatter on the slope. The effect of mouth width and groove center.

Example 2

如图7(a)所示。依次计算相邻两位置点x坐标的值差，保留零差值对应的位置点，形成坡口边缘预处理位置点集合

剔除差值不为零位置点集合

中对应的数据，分别如图7(b)中圆空心点和方实心点所示。通过由滤波阈值变量M_k形成的右边缘方向滤波器，即x_k≥M_k的高通滤波器，剔除远离中值的差值不为零位置点集合

中的位置点，获取坡口边缘保留位置点集合

如图7(c)中圆实心点和圆空心点所示；此时判断坡口边缘保留位置点集合

中数据极差R_k，该极差值R_k满足分散度判据R_k≤R_t，因此将坡口边缘保留位置点集合

作为坡口边缘真实位置点集合

如图7(d)所示。此时，因为坡口边缘真实位置点集合

中的位置点个数N_s大于INT(h×10％)，所以采用线性拟合的方法，获得坡口边缘重构位置点集合

该坡口边缘能反映实际的左坡口边缘形态，如图7(e) 所示。Fig. 7 is an embodiment of the global pattern recognition algorithm for the distribution of raw data on the right edge of the groove during pulse welding. Because the right edge 5 of the groove extracted in Fig. 6(i) above contains welding spatter, it is necessary to eliminate the influence of welding spatter on the edge position of the groove through the proposed outlier data filtering algorithm based on global pattern recognition. For the set of original position points of the groove edge composed of h data obtained from the right edge of the groove 5

As shown in Figure 7(a). Sequentially calculate the value difference of the x coordinates of two adjacent position points, retain the position points corresponding to the zero difference value, and form a set of preprocessing position points on the edge of the groove

Eliminate the set of points where the difference is not zero

The corresponding data in , are respectively shown in the circle hollow point and the square solid point in Fig. 7(b). Through the right edge direction filter formed by the filter threshold variable M _k , that is, the high-pass filter of x _k ≥ M _k , the set of points whose difference far away from the median value is not zero is eliminated

The location points in , get the collection of reserved location points on the edge of the groove

As shown in Figure 7(c) the circle solid point and the circle hollow point; at this time, it is judged that the edge of the groove retains the position point set

In the data range R _k , the range value R _k satisfies the dispersion criterion R _k ≤ R _t , so the set of position points on the edge of the groove is reserved

As a set of real position points on the edge of the groove

As shown in Figure 7(d). At this time, because the set of real position points on the groove edge

The number of position points N _s in is greater than INT(h×10%), so the linear fitting method is used to obtain the set of reconstructed position points on the edge of the groove

The groove edge can reflect the actual shape of the left groove edge, as shown in Fig. 7(e).

It can be seen from Figure 7 that the outlier data containing spatter in the figure is eliminated, and the retained real data can reasonably represent the actual groove edge distribution; it shows that the visual sensing detection method of narrow gap welding groove edge by global pattern recognition is effective. Higher detection accuracy, better anti-interference ability and stronger welding environment adaptability. By the same method as the groove right edge 5, the groove left edge 4 is reconstructed, and the position points of the reconstructed groove right edge 5 and the reconstructed groove left edge 4 are fitted to After the same welding area, the average value is calculated for the corresponding position, and the obtained groove width and groove center position sampling values are 11.71mm and 9.39mm respectively, and the detection accuracy is higher than the sampling value when the right edge 5 of the groove is not reconstructed.

Example 3

Figure 8 shows an example of image processing for narrow gap welding during DC welding. The test conditions include: after the signal trigger 9 receives the shaking arc position signal P _arc of the arc 2 at the side wall of the groove (the base value current signal _ib is not involved at this time), the shaking arc position signal P _arc is sent to the infrared The camera 8 starts to collect the global image 13 of the welding area. The shooting angle of the infrared camera 8 is θ=20°, the aperture is 16, and the exposure time is 3.5ms. The filter system 7 includes a UV mirror, a neutral filter with a transmittance of 10%, and a narrow-band filter with a center wavelength of 970nm and a bandwidth of 20nm. The base material of the welding specimen is Q370qE steel plate, the gap of welding groove 6 is 13.8mm; the current of arc 2 is 295A, the arc voltage is 27.6V, the welding speed V _w is 204mm/min, the dry elongation of welding wire is 20mm, and the welding wire The diameter of 3d is 1.2mm, the flow rate of welding shielding gas Ar-20% _CO2 is 25L/min; the arc-shaped shaking frequency of arc 2 is 2.5Hz, the shaking angle is 72°, and the residence time of both sides of the groove is equal. 100ms.

Based on the above test system and conditions, the global image 13 of the DC welding area where the arc 2 is located at the left edge 4 and the right edge 5 of the groove is respectively collected, and the image size is 544×544 pixels, as shown in Figure 8(a) and 8(b). Show. Among them, in Fig. 8(a), the arc radiation from the arc 2 to the left edge 4 of the groove is relatively serious, and at the same time, it can be observed that there is a welding spatter on the right edge 5 of the groove. By the same image processing method as the pulse welding image above, the extracted arc 2 region images are shown in Fig. 8(c) and 8(d), respectively. The ROI window of the groove edge position through position adaptation, as shown in the white box in Figure 8(e) and 8(f); the captured ROI window image of the groove edge, after local image processing of the groove edge, is obtained The groove edge image and the groove edge line are shown in Fig. 8(g) ~ 8(j).

It can be seen from Figure 8(i) that there is welding spatter on the right edge 5 of the groove extracted from the ROI window image of the groove edge, so an effective method must be adopted in the subsequent groove edge detection to avoid welding spatter The influence on the position value of the edge of the groove, thereby improving the detection accuracy of the groove width and the groove center.

Example 4

如图9(a)所示。通过依次计算相邻两个位置点的数值差，保留零差值对应的位置点，形成坡口边缘预处理位置点集合

剔除差值不为零位置点集合

中的位置点，分别如图9(b)中圆空心点和方实心点所示。因为坡口边缘预处理位置点集合

中的位置点不满足分散度判据R_k≤R_t，所以需要对坡口边缘预处理位置点集合

中的位置点通过由滤波阈值变量M_k形成的右边缘方向滤波器，即x_k≥M_k的高通滤波器，第一次剔除远离滤波阈值变量M_k的数据集合

获取一次保留位置点集合

如图9(c)中圆实心点和圆空心点所示；通过计算一次保留位置点集合

中的保留位置点的极差R_k后，此时位置点极差R_k不满足R_k≤R_t，所以需要通过高通滤波器进行第二次剔除远离中值的数据集合

获取二次保留位置点集合

如图9(d)中圆实心和圆空心所示；然后通过计算二次保留位置点集合

中的保留位置点的极差R_k，得到此时的R_k<R_t，则二次保留位置点集合

作为坡口边缘真实位置点集合

如图9(e)所示。通过统计坡口边缘真实位置点集合

中的位置点个数N_s，因为满足N_s≥INT(h×10％)，所以采用线性拟合方法，获得重构的坡口边缘重构位置点集合

该坡口边缘能反映实际的右坡口边缘形态，如图7(f)所示。通过重建的坡口右边缘和左边缘对应位置计算均值，获取的坡口宽度和坡口中心位置采样值分别为13.78mm和12.03mm，检测精度明显高于未对坡口右边缘5进行重建时的采样值。Fig. 9 shows an embodiment of the global pattern recognition algorithm for the distribution of raw data on the right edge of the groove during DC welding. Similar to the above-mentioned pulse welding method, the original position point set of the groove edge consisting of h data obtained for the right edge 5 of the groove extracted in Fig. 8(i) above

As shown in Figure 9(a). By sequentially calculating the numerical difference between two adjacent position points and retaining the position points corresponding to the zero difference value, a set of preprocessing position points for the edge of the groove is formed

Eliminate the set of points where the difference is not zero

The position points in , are respectively shown in the circle hollow point and the square solid point in Fig. 9(b). Because the groove edge preprocesses the location point set

The position points in do not satisfy the dispersion criterion R _k ≤ R _t , so it is necessary to preprocess the position point set on the edge of the groove

The position points in pass through the right edge direction filter formed by the filtering threshold variable M _k , that is, the high-pass filter of x _k ≥ M _k , and the data set far away from the filtering threshold variable M _k is eliminated for the first time

Get a collection of reserved location points

As shown in the circle solid point and the circle hollow point in Figure 9(c); by calculating once to reserve the position point set

After retaining the range R k of the position point in , the range R _k of the position point does not satisfy R _k ≤ R _t at this _time , so the data set far away from the median needs to be eliminated for the second time through a high-pass filter

Get the set of secondary reserved position points

As shown in the solid circle and hollow circle in Figure 9(d); and then by calculating the secondary set of reserved position points

The range R _k of the reserved position points in , get R _k < R _t at this time, then the secondary reserved position point set

As a set of real position points on the edge of the groove

As shown in Figure 9(e). Through the statistics of the real position point set of the edge of the groove

The number of position points N _s in , because N _s ≥ INT(h×10%) is satisfied, so the linear fitting method is used to obtain the reconstructed groove edge reconstruction position point set

The groove edge can reflect the actual right groove edge shape, as shown in Fig. 7(f). Calculate the mean value by calculating the corresponding positions of the right edge and the left edge of the reconstructed groove, and the obtained groove width and groove center position sampling values are 13.78mm and 12.03mm respectively, and the detection accuracy is significantly higher than when the right edge 5 of the groove is not reconstructed The sampling value of .

It can be seen from Figure 9 that the outlier data containing splashes on the right groove edge 5 is eliminated, and the retained real position point data can reasonably represent the actual groove edge distribution; it shows that the outlier data filtering algorithm of global pattern recognition It has higher detection accuracy, better anti-interference ability and stronger welding environment adaptability.

The above are only part of the specific implementation manners of the present invention. Certainly, the present invention also can have other multiple embodiments, without departing from the spirit and essence of the present invention, any person familiar with the technical field can make various corresponding equivalent changes and deformations according to the present invention , should belong to the scope of protection of the appended claims of the present invention.

Claims (10)

1. A visual sensor detection method for narrow gap welding groove edge based on global pattern recognition, based on the detection system including: infrared camera (8), filter system (7), signal trigger (9), image transmission data line (10) and a computer image processing system (11), wherein the signal trigger (9) is connected with the infrared camera (8), and the filter system (7) is coaxially installed on the infrared camera (8) On the lens, the computer image processing system (11) is connected with the infrared camera (8) through the image transmission data line (10); through the infrared camera (8), filter system (7) and signal trigger ( 9), collecting the overall image of the narrow gap welding area (13); it is characterized in that, the detection method specifically includes the following steps: 1) Determine the initial lateral position of the origin of the ROI window on the edge of the groove: the computer image processing system (11) extracts the position of the arc (2) highest point (14) of the arc from the global image (13) through global image processing, and determines the energy Capture the position of the SROI window (16) on the same side of the groove edge of the arc (2), or the position of the BROI window (17) that can capture both edges of the groove at the same time; by presetting the image captured by the SROI window (16) Image processing, obtaining the initial groove edge position on the same side of the arc (2), or by performing preset window image processing on the image captured by the BROI window (17), obtaining the initial groove edge on the same side and the opposite side of the arc (2) Position, to determine the initial lateral position of the origin (15) of the ROI window (12) on the opposite side of the groove edge of the arc (2) adaptively; 2) Obtain the set of original position points on the groove edge

After the position of the groove edge ROI window (12) is adaptively determined according to the position of the highest point of the arc and the initial lateral position of the origin (15) of the groove edge ROI window (12), the groove edge ROI window determined by the adaptive positioning method (12), intercept the ROI window image of the left edge of the groove (4) or the right edge of the groove (5) on the opposite side of the arc (2), and extract the left edge of the groove (4) or the right edge of the groove through local image processing (5), obtain the original position point set of groove edge composed of h data

Wherein, i represents current operation, and h represents the height value of ROI window (12); 3) Set the original position points of the obtained groove edge

The position points in are subjected to a data filtering: along the height h direction of the groove edge ROI window (12), the original position point set of the groove edge is calculated sequentially

The difference between the x-coordinates of two adjacent position points in , and the position points with a difference of zero are divided into the same position point subset; correspondingly, p(1≤p<h) position point subsets

is found, thus forming a groove edge preprocessing position point set containing p position point subsets

Correspondingly, p vertical line segments are formed, and each vertical line segment contains the same x-coordinate value, that is, the abscissa value; 4) Preprocessing the position point set on the edge of the groove

Perform secondary data filtering on the position points in : specifically include the following steps: ① Judging the set of preprocessing position points on the edge of the groove

Dispersion of the data in: the set of preprocessing position points for the groove edge obtained in step 3)

Firstly, determine the set of preprocessing position points on the edge of the groove

Dispersion of mid-position points; specifically: search for the set of pre-processing position points on the edge of the groove

The abscissa values of all position points in

The maximum value x _{k_max} and the minimum value x _{k_min} , calculate the range R _k ＝x _{k_max} -x _{k_min} ; judge the size of the range R _k and the data dispersion threshold R _t , the data dispersion threshold R _t = INT(h×4 %), when satisfying R _k ≤ R _t , the groove edge preprocessing position point set

As the set of real position points of the final groove edge

When R _k > R _t , the original position points of the groove edge are collected

The abscissa values of the h position points

median of

and

As the filtering threshold variable M _k , combined with the direction filter, after filtering outlier data, obtain the set of reserved position points on the edge of the groove

②Reserve the set of position points for the edge of the groove obtained in step ①

Judging the degree of dispersion of position points: by calculating the edge of the groove to retain the set of position points

The abscissa value of each position point in

Range R _k , when satisfying R _k ≤ R _t , the groove edge retains the set of position points

as the set of real position points G _i ^(s) of the groove edge; otherwise, calculate the set of retained position points of the groove edge

The abscissa value of the middle position point

median of

and

As the filtering threshold variable M _k ; through the direction filter, obtain the filtered groove edge reserved position point set

Repeat this process until

The extreme difference R _k of the abscissa value of each position point satisfies R _k ≤ R _t , forming a set of true position points G _i ^(s) on the edge of the groove; 5) Reconstructing the groove edge line: for the set of real position points G _i ^(s) of the groove edge obtained in step 4), count the number N _s of the real position points retained; when N _s ≤ INT(h×10% ), use the mean value calculation method to reconstruct the groove edge reconstruction location point set G _i ^(r) containing h groove edge location points; otherwise, use the least squares linear fitting method to reconstruct the Groove edge reconstruction position point set G _i ^(r) of the groove edge position points; where, the abscissa of the groove edge position point is

6) Obtaining the detection value of groove edge position: For the groove edge reconstruction position point set G _i ^(r) obtained in step 5) containing h groove edge position points, by calculating the abscissa of h groove edge position points The average value or fitting value of the groove edge position is obtained to obtain the sampling value of the groove edge position; the digital filtering method is adopted for the groove edge position sampling value to obtain the groove edge position detection value; 7) Repeat the above steps 2) to 6) until the welding process ends. 2. the narrow-gap welding groove edge visual sensing detection method of global pattern recognition according to claim 1 is characterized in that: in step 1), the specific content and method steps of described global image processing are: first to The global image (13) uses a median filter to perform image denoising; and then uses histogram analysis to calculate the frequency of pixels in different gray values in the global image (13), and then uses a global threshold to binarize the global image (13) ; Finally, after extracting the contour of the arc (2) through morphological operations, the coordinate value of the highest point of the arc (14) is obtained through grayscale search. 3. the narrow gap welding groove edge visual sensing detection method of global pattern recognition according to claim 1 is characterized in that: in step 1), the specific content and method steps of described preset window image processing are: Perform median filter denoising on the intercepted window image, contrast stretching to improve image contrast, apply Otsu threshold value processing to the window image, and then remove isolated points through morphological operations, and then use the Canny edge operator to extract the slope in the preset window image Mouth edge. 4. the narrow-gap welding groove edge visual sensing detection method of global pattern recognition according to claim 1 is characterized in that: in step 1), the longitudinal position of SROI window (16) and BROI window (17) is Determined by the ordinate value of the highest point of the arc (14); the abscissa position of the SROI window (16) is through the abscissa value of the highest point of the arc (14) and the abscissa value of the highest point of the arc (14) to the groove edge Set the distance to get. 5. the narrow-gap welding groove edge visual sensing detection method of global pattern recognition according to claim 1 is characterized in that: in step 2), the adaptive positioning method of the groove edge ROI window (12) The specific steps are: ① From the global image (13) of the latest N ₁ consecutive frames, extract the ordinate values of N ₁ arc highest points (14), and then obtain the filter value F ₁ of the N ₁ ordinate values through a digital filtering method, And take (F ₁ -δ ₁ ) as the ordinate value of the origin (15) of the groove edge ROI window (12); wherein, δ ₁ is the ordinate of the groove edge ROI window (12) origin (15) The correction constant of the coordinate position value, its value range is [-h,h]; ② Extract N ₂ groove left edge (4) or groove right edge (5) images from the recent N ₂ previous frame groove edge ROI window (12) images on the same side as the current groove edge to be detected Position value, then obtain the filter value F ₂ of its N ₂ position values by digital filtering method, and use (F ₂ -δ ₂ ) as the abscissa value of the origin (15) of the ROI window (12) on the edge of the groove; Or for the current global image (13), extract the current position values of the left edge of the groove (4) and the right edge of the groove (5) by the BROI window (17) that can intercept the double edges of the groove simultaneously, and then pass through digital filtering The method obtains the filter value F ₃ of the groove edge position value of the nearest N ₃ opposite sides of the arc, and uses (F ₃ -δ ₂ ) as the abscissa value of the origin (15) of the groove edge ROI window (12); wherein , δ ₂ is the correction constant of the abscissa position value of the origin (15) of the groove edge ROI window (12), which is the half-width of the groove edge ROI window (12), that is, δ ₂ =w/2, Wherein w is the width of the groove edge ROI window (12). 6. the narrow-gap welding groove edge visual sensing detection method of global pattern recognition according to claim 1, is characterized in that, in step 2): described local image processing, comprises the groove edge ROI window (12 ) image is subjected to median filter denoising, then contrast stretching is performed to improve the image contrast of the ROI window (12) on the edge of the groove, and then the partition threshold is used to obtain the binary image, and after further denoising by morphological operations, the ROI of the edge of the groove is extracted The edge line of the window (12) image. 7. The narrow-gap welding groove edge visual sensor detection method according to claim 6, characterized in that, the contrast stretching selects a segmented linear change function, and the morphological operation selects a closed operation, The edge extraction selects the Canny operator, and the partition threshold is to divide the groove edge ROI window (12) image into 4 equal parts on average, and then use the Otsu method threshold for the image of each partition respectively. 8. The narrow-gap welding groove edge visual sensor detection method according to claim 1, characterized in that, in step 4): the direction filter is expressed as: when the groove edge position When the point is on the left edge, the direction filter is a low-pass filter, expressed as x _{k ≤} M _k , that is, the position point whose abscissa value of the position point is less than or equal to the filter threshold variable M _k is reserved; when the groove edge position point When located on the right edge, the direction filter is a high-pass filter, expressed as x _k ≥ M _k , that is, the position point whose abscissa value of the reserved position point is greater than or equal to the filter threshold variable M _k ; x _k represents the circular data variable, and its value Equal to the set of groove edge preprocessing position points

or bevel edge retaining location point set

The abscissa value of the location point in . 9. The narrow-gap welding groove edge visual sensor detection method according to claim 1, characterized in that: in step 6), the digital filtering method is a limiter anti-pulse mean value filter, specifically comprising Follow the steps below: ① Determine the number of data N _f included in the filtering window: the filtering window includes the latest N _f sampling values of the edge position of the groove including the sampling value G _s of the edge position of the groove, and the filtering window includes the number of data N _f >2; ② Calculate the current sampling deviation η ₁ : Calculate the average value E _s of the sampling values of the N _f groove edge positions within the current filtering window, and calculate the absolute value of the difference between the current groove edge position sampling values G _s and the average value E _s , as this sampling deviation η ₁ ; ③Calculate this sampling deviation degree d _e : calculate the ratio of this sampling deviation η ₁ to the previous sampling deviation η ₀ , and use this ratio as this sampling deviation degree d _e ; ④ Repair abnormal sampling value: when the sampling deviation d _e is greater than the abnormal repair threshold d _th , the previous detection value of the edge position of the groove is used as the sampling value G _s of the edge position of the groove this time to realize the abnormal sampling of this time Value repair, the abnormal repair threshold d _th =3～10; ⑤Calculate the detection value G _d of the groove edge position this time: after removing the maximum value and the minimum value from the N _f groove edge position sampling values, calculate the mean value for the remaining (N _f -2) groove edge position sampling values , and use this mean value as the groove edge position detection value G _d this time. 10. An application of the narrow gap welding groove edge visual sensing detection method according to claim 1, characterized in that: it is used for the detection of the welding groove width, the groove center position and the welding process. Real-time tracking control; the applied method is, based on the detected left edge (4) and right edge (5) of the welded groove, by calculating the mean value of the corresponding positions of the two edges, the center of the welded groove is obtained; by calculating the corresponding position of the two edges Position difference, get the welding groove width value.

Images