TW550391B - Inspecting the defects on the conducting paths of printed circuit board (PCB) by using 2-D wavelet transform - Google Patents

Abstract

The aim of this research is to develop an approach for pre-selecting the serious defect candidates such as mousebite, spur, open and short on PCB conduct paths based on 2-D wavelet transform. According to the 2-D wavelet coefficients on each decomposition level and their inherent relationship between adjacent levels, the potential defects on conducting paths of PCB can be located under appropriate inter-level ratio (alpha), decomposition level, and wavelet basis.

Description

550391 92 · 4 · Under meal Λ EJ fir

Complement No. 5. Description of the invention (1) I. Background of the invention In recent years, machine vision technology has been widely used in the detection of defects in printed circuit boards. The calculation of detecting defects in printed circuit boards has many theoretical studies and The practical application system can be roughly divided into the reference method i ^ test method and hybrid method, and the above method of detecting printed circuit boards ^ a large number of models to compare or judge whether it meets a lot of specifications and conditions /, = time and vulnerable to image noise Impact, so printed circuit board manufacturers need a fast and effective automatic visual inspection system. At present, lightweight and exquisite and functional electronic products have become a trend. &Amp; Micro size, fine line width and fine brush circuit boards have emerged at the historic moment, making printed circuit board circuit defects: difficult and easy to appear. Fine line width and brushing in the industry at present: Printed circuit

Uous uses ite), line highlights, and short cuts (see section: 闰 ^ ^ a ugly road JT, Guang > Mang Chenggu 1) as uncommon and serious. Printed circuit board line 踗 =: If there are the above defects, the printed circuit will not generate a road algorithm to reduce the misjudgment mechanism: even the effective flaw detection II. Summary of the invention The present invention uses a pre-selected printed circuit with a network and a broken circuit Board line… Short circuit defects, especially thin line width and fine line 550391 V / day repair ^ V, invention 0) 2) BGA substrate "Refer to the second conversion can be in λ W ~ situation Λ Xing shoulder pair of pumps. Two-dimensional The wavelet part will capture two: f multi-level decomposition into two parts of smoothness and detail, three parts of vertical details :::: area; mutation and decomposition into the diagonal, horizontal and straight parts The level of the detail line is not equal to that of the vertical line of the gate, so you can use this feature to calculate the energy clustering ratio of the pixels on the edge of the line.

Hi value range㈣ This month, we proposed to use the clustering wavelet energy% — range, the recommended wavelet /, and the range of sigh α values to pre-select the fatigue fatigue i = 2D wavelet transform order of printed circuit board detection The method has real flaws. Traditionally, the present invention can reduce the detection range. Again = candidate pixels for detection, feature comparison and calculation that need to be referenced. Shiming does not need training samples, and it does not have a variety of printed circuits: the influence of the light source is not obvious, so the system can be detected. I 轾 h I. A fast and effective automatic vision II. Detailed description of the invention The main task of the invention is to use two-dimensional small lines to highlight, _ & wave conversion pre-selection printing open circuit, short circuit and other defects. Kushiro; therefore, the obvious defects in the area are the strong skin transformation of the horizontal and vertical part of the Selangor order. V: Eight are presented in two-dimensional Xiaofen, its principle and two-dimensional wavelet, page 6 550391. 92. 4 · ⑽Amendment on 8th of August, supplementation_5. Description of the invention (3) The conversion to remove image noise is similar. The two-dimensional wavelet is composed of a one-dimensional horizontal and vertical common function p and a generating function 0, that is: Φ (χ, y) Ψν (χ, y) ΨΗχ, y) Ψά (χ, y) where Φ (χ , y) Ψν (χ, y) ΨΗχ, y) Ψά (χ, y) < (χ) φγ (χ) 0h (χ) 0ν (χ) = ph (x) X φγ (γ) = 0h (x ) χ φγ (γ) = ph (x) χ 0v (y) = 0h (x) χ 0v (y): 2D wavelet common function: 2D vertical wavelet mother function: 2D horizontal wavelet mother function: 2D pair Angular wavelet mother function One-dimensional horizontal wavelet common function One-dimensional vertical wavelet common function One-dimensional horizontal wavelet mother function One-dimensional vertical wavelet mother function Therefore, the details will be further divided into horizontal, vertical and diagonal parts. Among them, Φ (χ, y) captures the smooth part of the input image; and ¥ h (χ, y), Ψν (χ, y), and ¥ d (x, y) capture the detail level, detail vertical, and detail of the input image, respectively Diagonal section. If an input image F (χ, y) is an image of m χ η

(5) Invention description (4) The prime matrix is composed, then F (X, y) can be approximated by the above two-dimensional wavelet ', that is, φ ^ (^ 7) + ΣΣ ^ «ψν (κ, 7) + ΣΣ ^ ν ”J4 mil where ^ ll Φ J, M, M (X, y) F (x, y) dxdy d ^ JJ ψΤΓι7 ·, 3«, M (χ? Γ) P (χ, y) dxdy ^ Κλ «, »« JJ Ψκ .; >, »(x, y) F (x, y) dxdy ^ y, w,« ^ JJ Ψά.7 > ^ (x? Y) F (x, y) dxdy level J Wavelet coefficient matrix of smooth part (m χ η): Wavelet coefficient matrix (m χ ^) of vertical part of j-th order details (j = 1, 2, ..., J) Wavelet coefficient matrix (m χ n) (J · 1,2,2, ..., J) Therefore, the second slip, the first is written, and the fine: the wavelet coefficient matrix of the diagonal part of the jth order detail (bu 1, 2, ... , J) 'Dimensional wavelet transform can calculate the smooth wavelet coefficient Λ, which is the angle corresponding to every detail like the horizontal and vertical details of the fish; ^ Small: the first abbreviation of the vertical wavelet number in the leather system is abbreviated as γ Factory d and details 550391 1992. 6 \) Ά θ Revision supplement V. Description of the invention (5) Secondary image of wavelet coefficient (abbreviated as ~ —S. + 1, S-d and d are factories ”j_Sj ⑷ Decompose and purchase eight red J + 1 dj + 1 j + 1 ~ sj + i and dchuan—, the fourth picture illustrates this: the knife solution program. The fifth picture is a "x_b〇x", the picture is transformed by two-dimensional wavelet The later result (J = 2, wavelet base: d4). The sixth figure is the result of a BGA substrate line edge processing after two-dimensional wavelet transformation (J = 2, wavelet base: d4), among which the whiter pixels The point has a larger absolute value of the wavelet coefficient.

Two-dimensional wavelet transform is often used to remove image noise. Its principle is to use the exponential relationship between the wavelet coefficients between the jth order and the j + 1th order after multi-level decomposition of the two-dimensional wavelet transform; and then to distinguish that this pixel belongs to the edge of a normal image Or image noise. For a pixel (xQ, yG), the two-dimensional wavelet energy ((x0, y.)) In the jth order is defined as:

Aj (X0, y0) = tan Ί j (X〇, y〇)) d j (x〇, y〇)

Where dS (xG, y) ··· pixel (X., yG) is transformed into a two-dimensional wavelet coefficient at the jth order via wavelet Ψ1 (X, y) (Yl (x, y) = OTh (x, y) / ax)

550391 9 | 4Λ30Β? Duozheng supplement V. Description of the invention (6) dj (X〇, y.): The pixel (XG, yQ) is transformed by the wavelet ψ2 (χ, y) in the j-th order wavelet coefficient ( ψ2 (χ 30 = / to do)

Aj (X0, yG): the direction of the gradient vector of the pixel (Xq, y.), That is, the direction of the gradient vector of / σ pixel (XG,), searches for the two-dimensional wavelet coefficient with the greatest variation. For a certain pixel ( %, YQ) The two-dimensional wavelet grouping energy (N〆%, yQ)) in the first order is defined as:

Nj (x〇 ^ 〇) = where

Ds = Kx? Υ) I (x-X〇) 2+ + (y-Y〇) 2 ^ Ks2, (y-y〇) / (x-X〇) tan'1 (AjCxo, y〇))}

K s wavelet mother function support length (half of support length) (example: half of the support length of wavelet basis d4 is 2) (example: half of the support length of wavelet basis d 6 is 3) (j = 1, 2, ·· ·, J)

Page 10 Remedy V. Description of Invention (7)

Furthermore, if the pixel (χ) A wave group gathers energy (N. + 1 (xv ^ ^ shirt-like noise, its j + 1 卩 卩 ai, J 0, y〇)) divided by 笫 Ί N is two-dimensionally smaller than 々—1 (J (X0, y °)) must be greater than or equal to 2. That is, the dimensional wavelet grouping energy (N; j + 1 (x〇5 y〇) / Nj (x〇, y〇))-2 ail its application ^ ·· The wavelet energy grouping ratio is therefore 'α value is greater than or equal to 〇 Catch, the image noise can be removed. The prime phase is collected and the image is reprinted. It is a strong print: 1 circuit board circuit gap, line protrusion, open circuit, short circuit, etc. IT is wide: strong 2 irregular lines; therefore, the image value of the defect a The range of the clustering ratio between normal and normal line waves includes missing π and: values are not good. As long as you find the correct a, you can greatly reduce the number of defective pixels such as protruding roads, open circuits, and short circuits. The irregular circuit of the printed circuit board The present invention defines an image = concentrating energy to comply. The testing environment of the printed circuit board, the original y /)) is as follows: ', Xl' yi <)) The two-dimensional wavelet energy at the Jth order (WM ^ x /, WMiU ”^ ί °) = + | < ^ (Ύ) | 2

It ’s called food. It ’s called the fifth. Invention description (8) Lizhong / d'.h0, V: Pixels (x /, y /) are two d '· (χΛ in the vertical part of the j-th dimension wavelet coefficient detail y /) ·· The pixel point (χΛ V) is at μ in the jth order, the dimension wavelet is mixed to the desired pixel point (XiG, y,) in the horizontal part to define its clustering range (Drj ), The present invention defines the group-wave group-focusing energy; the range of pre-focusing needs to be as follows: pj = {(x, y) llx-xf | 2 < τ 52, | yy · | 2 < τ ε2} where r = wavelet mother function support Length (SUppoj ^ leDgthjs two 2j (bu 1, 2, · · ·, j) half of the clustering range defined by the present invention (a square pixel window with the search as the center,) a pixel (χΛγι〇) ·· ( 2⑽ [fxs])) + 1, and the number of pixels contained by Qiong Chang is as the prime wavelet grouping energy of prime 黠 (X /, y /) in the first order. This square is like 550391 92 4. 3U 修 止 Λ 曰 4, Complement No. 5. Description of the invention (9) Calculated in the window of the prime point, the method defined by the present invention is as follows: :: Pixel point (χΛ 为 ι) as the center, find its eight pixels with the largest two-dimensional wavelet energy (definition 1 'prime A pixel with a large two-dimensional wavelet energy (defined as (x 二 // ι)) and a secondary oi of zero XC is 0J W. It records its order 2 and the latter 3. Take the pixel (Xil, y ") as The center, find its 8 pixels with the largest two-dimensional wavelet energy (defined as the pixel point (Xi-1, y "l) in the occupied pixel point in the% bearing point) as the center, find its 8 = (Xi , Yi2)), with the pixel point with the largest two-dimensional wavelet energy (defined as (V2, y 「2)). Record WMj (Xi2, yi2) and WMj (V2, y「 2) When the value is searched, the search direction is carried out in two directions. 7 They are 0. 5 Repeat steps 3, 4 to get WMjXl3, yi3) and WM] (x, one side stops when it crosses the window boundary.) 1 ..., if any When both sides exceed the window boundary, the search stops and the pixel (XiG, y /) is two-dimensionally small (WNjUi., Y /) at the jth order as follows:

V. Description of the invention (10) id: sunny j (x 八 yi) 8 · Recovering records that have been recorded below zero _ -4 * #, two-dimensional + of Beijing, next iteration 'skin energy The value of Levi's seventh figure is an example of calculating the wave group concentrating energy. Through the two-dimensional wavelet group mouth defined by the present invention, a line protruding, open circuit, short circuit and other defects, such as turtles, are used to capture defects.) If it is a defective pixel, then its 雉., ~ Pixels (0 its j-th order two-dimensional wavelet grouping energy small energy grouping ratio (α) is less than zero. 0 the following formula), that is, its wavelet

(WN j + l

Xi ° iy1 °) / WNJ (Xi〇, yi〇)) = 2 α + ΐ < 2, 1 a < 0 However, which wavelet base is used, (j • value) can fully capture the real Defects such as the two-dimensional wavelet transformation hierarchy (at this time, there are important points in this line & ^ line highlights, open circuits, short circuits. The experimental sample is: the Λ road turn has also been captured) There are also about 330 X 350 images of this image 像Real BGA substrate; each BGA substrate shadow has 5: n cataclys (about 40 pixels / mm) per contact substrate i), short-circuited person ..., two: yes. Each line is prominent and 5 open? Ai ... a total of 60 manual defects (see figure 8). Make page 14 550391 92. 4. όυ amend year Λ. Shi Buwuwu. Description of the invention (11) Using d4, s4, d6, s6, c6, d8s8f (b. 1, 2) has a better detection area. The second-order two-dimensional small chirp, target force, so the present invention uses (α), the experimental results obtained, = calculated wavelet energy clustering ratio name, as shown in the main table. In the table, although there are hundreds of pixels (-pixels ..., choose pixels), if the wavelet basis is used in the candidate pixels of Ran cyuanhou, each of the candidate pixels is included. Reality: Missing and /: The selected circuit is broken, the short gap seems to have a consistent gap, the line is prominent, and the error of "missing" with α less than ㈤ without. Moreover, if the line is highlighted, the real gaps, meanings, short circuits, etc. cannot be fully captured (see the second table) ’, so α is less than zero.

Righteousness is a valid range. 7 J P.15 550391 Private 4 · 3Q repair stop * Λ S supplement V. Description of the invention (12) The first table · Experimental results (α < 0)

CL < 0 j =! J = 2 mm Finding the head mt Recording & Finding the head a ?: Weime me ^ i) Pinaru d4 Transcript 1 318 5 10 370 1 1 Sample 2 272 4 593 0 Sample Three 372 1 54} 0 s4 sample one 318 5 10 370 1 1 sample two 272 4 593 0 sample three 37S 1 54] 0 c6 sample one 409 0 0 372 0 0 sample two 366 0 591 0 sample three 574 0 59B 0 άύ Sample 1 298 2 4 295 0 1 Sample 2 259 1 459 1 Sample 3 374 1 405 0 s6 Transcript 1 298 2 4 295 0 1 Sample 2 259 1 459 1 Sample 3 374 1 405 0 d8 Sample 1 19S 6 15 156 4 12 Sample 2 157 6 156 6 Sample 3 210 3 245 2 s8 Sample 1 225 5 S 235 0 5 Sample 2 189 2 271 5 Sample 3 310 1 389 0 Page 16 III ·· 550 391 92 · 4. Training Correction Said i > Bu Wuwu, description of the invention (13) α < -0 ^ j = 2 The teaching method of wavelet base dwellings was mistakenly read by ^^ 's Detective essay "1 the number of gastric cords Treasure _ dip method is guaranteed to detect a piece of mistakes and transcripts of compiling teachings (annoyances) and »Mu Qing box mistakes) 隹 和 Sample 1 172 7 272 2 Sample 2 159 7 19 542 2 6 Sample Book 3 230 5 472 2 Sample 1 172 7 272 2 s4 Sample 2 159 7 19 542 2 6 Sample 3 230 5 472 2 Sample 1 349 4 304 1 c6 Sample 2 169 2% 401 2 4 Sample 3 354 2 479 1 Transcript 1S9 3 348 1 d6 sample two 227 2 9 328 2 4 sample three 223 4 326 1 sample one 189 3 348 1 s6 sample two 227 2 9 328 2 4 copy three 223 4 326 1 copy one 113 7 126 6 d8 Sample 2 92 8 20 156 8 16 Transcript 3 124 5 199 2 Sample 1 131 6 194 1 s8 Sample 2 105 5 17 350 6 8 Sample 3 174 6 332 1 Table 2 · Experimental results (a $ -0.5) mill Page 17 丨 Supplement I Schematic description of the first picture: the traditional printed electricity, the anti-family area: 50mm X 40mm second picture: BGA substrate, field of view: & mm X 20mm second picture: common printing Circuit handle line protruding (C) Open circuit (d) t = Type of heavy defect: (a) Notch (b) Order decomposition procedure 4th figure: Two-dimensional wavelet transforms are much more Fifth figure: 7: Γ The graph passes through two-dimensional wavelets Transformed results (J = 2, wavelet 2 1 '· (a) X —b〇x " Figure (b) is transformed by two-dimensional wavelet The second /, Kh-BGA substrate circuit edge map results after two-dimensional wavelet transformation J 2, wavelet base: d4): (a) BGA substrate circuit edge map (b) results of each layer after two-dimensional wavelet transformation Figure 7: An example of calculating the two-dimensional wavelet grouping energy: ((a)-(g)) Figure \ Experimental sample of a real BGA substrate (already marginalized) (a) Sample 1 (b) Sample (C) Sample three

Page 18

Claims (1)

Sixth, the scope of patent application 1 ~! I use two-dimensional wavelet transform (2-D Wavelet Transform) to detect P brush circuit board gap (Mousebite), line protrusion (SpUr), open (Open) and short circuit (Sh. rt) and other circuit defects, including: finding the two-dimensional small amount and two-dimensional wavelet grouping energy of each pixel of the 2 ^ circuit board image at the first and second orders; and: The two-dimensional wavelet grouping energy ratio of the first order and the second order 2. The method according to item 丨 of the patent application range further includes: The two-dimensional wavelet used is a Coif lets wavelet, and its support length is 6. 3. The method according to item 丨 of the scope of patent application, which further includes the two-dimensional wavelet energy (WM) of the pixel point (X, y) in the "" th order "(χ, with j as the side slave y) = yP7 + | ^ (x, y) I2 wavelet system wavelet system 2 (x, y): 2D of the vertical part of the pixel (x, y) in the j-th order detail = (x, y) ·· pixel ( x, y) 2D at the level of detail of the jth order 4. The method of item 丨 in the scope of patent application, where the two-dimensional wavelet grouping energy (WNj (x, y)) of the image order is evaluated as /, x , Y) in the first] WNj (x, y) = full WMj (x Ά 550391 lawsuit 4.30 amendment date / supplementary six, the scope of the patent application where nl and n2 are the affected pixels in the clustering range (X, y ) Number of pixel points for two-dimensional wavelet grouping energy. ♦ 5. If the method of the first item in the scope of the patent application, the pixel point (X, y) first-order and second-order two-dimensional wavelet grouping energy ratio (α) The evaluation is: + ι WNl (x, y) WN2 (k, y) 6. As the method of the scope of patent application No. 5 further includes: if the two-dimensional wavelet grouping energy ratio of the pixel (X, y) is less than Zero, it may be a gap, line And other out, open or shorted defective pixels. Page 20