Report

Assessing the quality of spot welding electrode’s tip using digital image processing techniques A .A. Abdulhadi Coherent and Electro-Optics Research Group GERI Presentation headlines Resistance spot welding Effects of increased electrode’s diameter Assessing the quality of the electrode automatically Flat tip Doom tip Building a system that assess the quality of welding electrode automatically. Future work Resistance spot welding Resistance spot welding is a quick and easy way to join two materials Two electrodes are used to perform spot welding ; they are placed either side of the surfaces to be welded The functions of the two electrodes are 1) clamping of the work 2) applying the weld force required for welding 3) applying the weld current necessary for fusion of the work pieces 4) a final retraction of the electrodes after the molten nugget has solidified Effects of increased electrode diameter Diameter high, area high Resistance low Heat low Pressure lower Quality of welding nugget is worse original image 50 It is normal for the electrodes to wear to such excess that they need redressing, or replacing. 100 150 200 50 100 150 200 250 300 New electrode original image This wear varies according to the applied current and the material thickness. 50 100 150 200 50 100 150 200 250 after wearing 300 Assessing the quality of the electrode automatically We need a method to assess the quality of the welding electrode automatically. We capture an image of the electrode using a digital camera We process this image using digital image processing techniques to evaluate the quality of the electrode. Digital image processing techniques to assess the quality of welding electrode Extract the electrode from the image Image segmentation Determine the boundary for the electrode Filter this boundary using boundary representation and description methods Find the width of the tip using Cullen method If the width of the tip is smaller than a predefined threshold, we consider the tip as a good tip Otherwise the tip needs replacing or redressing Image segmentation There are many image segmentation methods Edge detection Sobel Canny Laplacian Prewitt .... Hough transform Region growing Graph theory Snakes active contours Electrodes types We have two types of electrodes Flat tip Doom tip For each tip type, we have a bank of 250 images. Images for tips with high quality Images for tips with low quality 20 40 60 80 100 120 140 160 180 200 220 50 100 150 200 250 300 original image 50 100 150 200 50 100 150 200 250 300 Cullen method The top Figure shows an image that contains a flat tip. The bottom figure shows a schematic diagram of an ideal flat tip and indicates its parameters such as the tip width Tp and the electrode width Cp. 20 40 60 80 100 120 140 160 180 200 220 50 100 150 200 250 300 Cullen method This figure show the boundary of the electrode. Let us process the boundary image on a row by row basis. For each row, we subtract the x coordinates of the left boundary points (shown in red colour) from the x coordinates of the right boundary points (shown in blue colour). Cullen method The first xa rows do not contain the tip boundary. The subtraction operation produces zeros as shown in the bottom Figure. For the row xa +1, the subtraction operation produces the tip width Tp. For the rows from xa + 1 until xa + xg, the subtraction operation produces a line with a slope of g Cullen method The slope of g =2 For the rows from xa + xg + 1 until M, the subtraction operation produces a value of Cp. The first derivative of the 2D top graph is calculated and this is shown in the bottom figure. Cullen method The width of the tip is determined as follows. The derivative of the tip profile is thresholded using a threshold value g. Then the number of points whose values are larger than g is determined and this number is assigned to xg. The tip width is then determined using the Equation 20 40 40 60 60 80 80 100 100 Pixels 20 120 120 140 140 160 160 180 180 200 200 220 220 50 100 150 200 250 300 50 100 150 Pixels 200 250 300 35 160 30 First derivative of the tip profile 140 120 Tip profile 100 80 60 40 25 20 15 10 5 20 0 0 -5 0 50 100 150 x axis 200 250 0 50 100 150 x axis 200 250 20 40 60 80 100 Assessing the quality of flat tips 120 140 160 180 200 220 50 100 150 200 250 300 Image segmentation and boundary representation Canny algorithm and Cullen method Tp found manually and Tp found using Canny and cullen 140 130 120 110 100 90 80 Tp manually Tp Canny and Cullen 70 0 50 100 150 Image Number 200 250 300 Tp manually and Tp Canny and Cullen Error between Tp found manually and Tp found using Canny and cullen 10 5 Tp Error obtained automatically for two hundreds and fifty images, we have used Canny algorithm for image segmentation, and Cullen method for extracting the tip width The results are shown in red The tip width has been determined manually for the 250 images and the results are shown in blue. The bottom figure shows the differences between the manual and automatic determination of the tip width in pixels. Tp Manually obtained and Canny and cullen To determine the width of the tip in pixels 0 -5 -10 -15 0 50 100 150 Image Number 200 Error between manually and automatically measurement the tip 250 300 Boundary filtering using Fourier transform Suppose that we have the ( x7 , y 7 ) boundary shown here ( x6 , y 6 ) x(n)= x(1)+x(2)+...+x(K-1) y(n)= y(1)+y(2)+...+y(K-1) We can represent the boundary using the complex ( x7 , y 7 ) numbers s(k)=x(1)+iy(1)+ x(2)+iy(2)+... x(K-1)+iy(K-1)+ i 1 ( x2 , y 2 ) ( x3 , y 3 ) ( x4 , y 4 ) ( x6 , y 6 ) ( x5 , y 5 ) The discrete Fourier transform of s(k) is The inverse Fourier transform of these coefficients restores s(k). That is, Suppose, however, that instead of all the Fourier coefficients, only the first P coefficients are used. This is equivalent to setting the term a(u) = 0 for u > P-1. Then we get an approximation for the boundary. The low frequency components account for the global shape of the boundary Whereas the high frequency components account for the fine details in the boundary shape Canny algorithm, Fourier transform Tp found manually and Tp found using Canny and Fourier transform 140 Tp Manually obtained and Canny and Fourier transform To determine the width of the tip in 130 pixels automatically for two hundreds and fifty images, we used Canny algorithm for image segmentation, Fourier transform for filtering the boundary Tp manually and Tp Canny and Fourier Cullen method for extracting the tip Transform width The results are shown in red The tip width has been determined manually in the 250 images and the results are shown in blue. The bottom figure shows the differences between the manual and automatic determination of the tip width in pixels. Error between manually and automatically 120 110 100 90 80 Tp manually Tp Canny and fourier transform 70 0 50 100 150 Image Number 200 250 300 Error between Tp found manually and Tp found using Canny and Fourier transform 10 Tp Error obtained 5 0 -5 -10 -15 0 50 measurement the tip 100 150 Image Number 200 250 300 Canny algorithm and minimum –perimeter polygons A closed boundary can be approximated can be approximated with arbitrary accuracy by a polygon. For a closed boundary, the approximation becomes exact when the number of vertices of the polygon is equal to the number of points in the boundary, and each vertex coincides with a point on the boundary. The details and the noise in the boundary can be reduced by decreasing the number of vertices. Canny algorithm and minimum –perimeter polygons To determine the width of the tip in pixels Tp Manually obtained and Canny and Polygon automatically for two hundreds and fifty images, we used Canny algorithm for image segmentation, Minimum perimeter polygon for filtering the boundary Tp found manually and Tp found using Canny and Polygon 140 130 120 110 100 90 80 70 Cullen method for extracting the tip Tp manually Tp Canny and polygon 0 50 100 150 Image Number 200 250 300 width Tp manually and Tp Canny and MP Polygons Error between Tp found manually and Tp found using Canny and Polygon 15 manually in the 250 images and the results are shown in blue. The bottom figure shows the differences between the manual and automatic determination of the tip width in pixels. 10 5 Tp Error obtained The results are shown in red The tip width has been determined 0 -5 -10 -15 0 50 100 150 Image Number 200 250 300 Error between manually and automatically measurement the tip Region growing algorithm and Cullen method The region growing is a procedure that groups pixels or sub-regions into larger regions based on predefined criteria for growth Starting with a single pixel (seed) and adding new pixels slowly 1- choose the pixel, 2- check the neighbouring pixels and add them to the region if they are similar to the seed, 3 – repeat step (2) for each of the newly added pixels; stop if no more pixels can be added Region growing algorithm and Cullen method To determine the width of the tip in pixels 130 Tp Manually obtained 120 110 100 90 80 70 0 50 100 150 Image Number 200 250 300 Tp manually and Tp Region growing and Cullen Error between Tp found manually and Tp found using region growing and Fourier transform 14 12 10 8 Tp Error obtained automatically for two hundreds and fifty images, we used Region growing for image segmentation, Cullen method for extracting the tip width The results are shown in red The tip width has been determined manually in the 250 images and the results are shown in blue. The bottom figure shows the differences between the manual and automatic determination of the tip width in pixels. Tp found manually and Tp found using region growing and Fourier transform 140 6 4 2 0 -2 -4 -6 0 50 100 150 Image Number 200 250 300 Error between manually and automatically measurement the tip Region growing Algorithm, Fourier Transform Tp found manually and Tp found using region growing and Fourier transform 140 Error between Tp found manually and Tp found using region growing and Fourier transform 25 20 130 15 Tp Error obtained Tp Manually obtained 120 110 100 10 5 90 0 80 70 -5 0 50 100 150 Image Number 200 250 Tp manually and Tp Region growing and F Transform 300 0 50 100 150 Image Number 200 250 300 Error between manually and automatically measurement the tip Region growing algorithm and minimum – perimeter polygons Error between Tp found manually and Tp found using region growing , Polygon 14 Tp found manually and Tp found using region growing, Polygon 12 130 10 120 Tp Error obtained Tp Manually obtained and region growing, Polygon 140 110 100 90 6 4 2 0 80 70 8 Tp manually Tp region growing and polygon 0 50 100 150 Image Number 200 250 300 Tp manually and Tp Region growing and polygon -2 0 50 100 150 Image Number 200 250 300 Error between manually and automatically measurement the tip The region growing and minimum perimeter polygon case is superior because the error term for Tp is smaller as shown in Figure above Graph theory algorithm and Cullen method Tp found manually and Tp found using Normalized cuts and cullen The set of points in arbitrary feature space are represented as a weighted undirected graph G (V , E ) Tp Manually obtained and Normalized cuts and cullen 150 140 130 120 110 100 90 80 70 50 100 150 Image Number 200 250 300 Tp manually and Tp Normalized Cuts and Cullen Error between Tp found manually and Tp found using Normalized cuts and cullen 15 10 5 Tp Error obtained where the nodes of the graph are the points in the feature space, and an edge is formed between every pair of nodes. Tp manually Tp normalized cuts and Cullen 0 0 -5 -10 -15 The weight on each edge w(i,j), is a function of the similarity between nodes i and j . -20 0 50 100 150 Image Number 200 250 300 Error between manually and automatically measurement the tip Which method is the best? we have calculated the standard deviation for error between the manual and automatic methods for the two hundreds and fifty electrode tip images. The results are shown in table. The results of this table reveal that the region growing and Minimum –Perimeter Polygons gave the most accurate method for determining the tip width. On the other hand, the graph image segmentation algorithm produces the worst results. Cases 1 Canny algorithm and Cullen method standard deviation 4.50 2 Canny Algorithm, Fourier Transform 4.80 3 Canny Algorithm and Minimum – Perimeter Polygons 4 Region grown Algorithm and Cullen method 5 Region grown Algorithm, Fourier Transform 6 Region grown Algorithm, and Minimum –Perimeter Polygons 7 Graph Theory Algorithm and Cullen method 4.70 3.80 3.70 3.40 5.60 Doom electrode tip The image for the doom electrode is very hard to segment. This is because of the shining parts of the tip doom. We show here the segmentation results using Laplacian Sobel Prewitt Canny None of these edge detection algorithms works properly Also, we have attempted these algorithms Region growing Graph theory Hough transform Also none of these edge detection algorithms works properly The only image segmentation algorithm we tried it and can segment the doom electrode successfully is the Active contours snake algorithm Laplacian Algorithm Original image Sobel Algorithm Prewitt Algorithm Canny Algorithm Snake Algorithm Snake are curves defined within an image domain that can move under the influence of internal forces coming from within the curve itself and external forces The internal and external forces are defined so that the snake will conform to an object boundary traditional snake is a curve X(s)=[x(s),y(s)], s [0,1] Finding a method for determining the diameter of the tip automatically. Image segmentation & representation Original image Thresholding Snake Algorithm Snake Algorithm Snake algorithm and Cullen Tp found manually and Tp found using Snake.Cullen Error between Tp found manually and Tp found using snake.ahmed 140 30 20 120 Tp Error obtained Tp Manually obtained 10 100 80 0 -10 60 -20 40 -30 Tp Manually obtained Tp Snake Cullen 20 0 50 100 150 Image Number 200 250 Tp manually and Tp Snake and Cullen 300 -40 0 50 100 150 Image Number 200 250 300 Error between manually and automatically measurement the tip Snake algorithm and Fourier transform Tp found manually and Tp found using Tp.Snak.Fourier Error between Tp found manually and Tp found using Tp.snak.Fourier 140 60 50 120 100 Tp Error obtained Tp Manually obtained 40 80 30 20 10 60 0 40 Tp Manually obtained Tp snak Fourier 20 0 50 100 150 Image Number 200 250 300 Tp manually and Fourier Transform -10 -20 0 50 100 150 Image Number 200 250 300 Error between manually and automatically Snake algorithm and polygon Tp found manually and Tp found using Tp.Snake. Polgyon Error between Tp found manually and Tp found using Tp .polgyon .snak 140 40 30 120 Tp Error obtained Tp Manually obtained 20 100 80 10 0 60 -10 40 Tp Manually obtained Tp Polgyon Snak 20 0 50 100 150 Image Number 200 250 300 Tp manually and Tp Snake and MP Polygons -20 -30 0 50 100 150 Image Number 200 250 300 Error between manually and automatically Cases 1 Sank algorithm and Cullen method standard deviation 7.3801 2 Sank Algorithm, Fourier Transform 6.8742 3 Sank Algorithm and Minimum – Perimeter Polygons 7.1366 we have calculated the standard deviation for error between the manual and automatic methods for the two hundreds and fifty electrode doom tip images. Built system that can assess the quality of spot welding electrodes easily We need to improve the performance of determining the tip width automatically. To do this We use a high performance illumination source to illuminate the electrode We capture an image for the shadow of the electrode The shadow image is easy to process thresholding Then we can extract the tip width easily using Cullen method Original image edge Thresholding edge Future work To embed the system that can assess the quality of spot welding electrodes into a spot welding machine. Conclusions We have used image processing algorithms successfully to assess the quality of spot welding electrodes automatically. We have built a system that can assess the quality of spot welding electrodes using simple image processing techniques Thresholding Simple filtering methods such as median filtering Any questions