Int J Adv Manuf Technol (2010) 51:995–1008 DOI 10.1007/s00170-010-2664-9

ORIGINAL ARTICLE

Misalignment inspection of multilayer PCBs with an automated X-ray machine vision system Shui-Fa Chuang & Wen-Tung Chang & Chih-Cheng Lin & Yeong-Shin Tarng

Received: 22 June 2009 / Accepted: 7 April 2010 / Published online: 1 May 2010 # Springer-Verlag London Limited 2010

Abstract In recent times, multilayer printed circuit boards (PCBs) have been extensively applied in the electronics industry owing to their high capacities for complex and densely packed circuit layouts arranged in a limited space. The inspection of fabricated multilayer PCBs is thus important in order to ensure quality control and improve the fabrication process. In this paper, an automated X-ray machine vision system was developed exclusively for the inspection of the layer-to-layer misalignment of laminated multilayer PCBs. Based on a mechatronics system and X-ray image processing techniques, an automated misalignment inspection process was established. Experiments meant to inspect three critical layer-to-layer misalignment modes, expansion, contraction, and offset, found within ten-layer PCB samples, were conducted to test the feasibility of the developed machine inspection system. The experimental results show that the developed X-ray machine vision system, combined with the automated misalignment inspection process, was able to undertake misalignment inspection of certain multilayer PCBs. Keywords Multilayer printed circuit board (PCB) . Layer-to-layer misalignment . Misalignment inspection . X-ray image . Machine vision . Image processing . Mechatronics system

Nomenclature A, B, C, D

Ai, Bi, Ci, Di Aj, Bj, Cj, Dj d(AB), d(BC), d(CD), d(DA)

di(AB), di(BC), di(CD), di(DA)

eij(A), eij(B), eij(C), eij(D)

OA, OB, OC, OD OCCD-XCCDYCCD Omp-XmpYmp

S.-F. Chuang : C.-C. Lin : Y.-S. Tarng Department of Mechanical Engineering, National Taiwan University of Science and Technology, Taipei 10607, Taiwan W.-T. Chang (*) Opto-Mechatronics Technology Center, National Taiwan University of Science and Technology, Taipei 10607, Taiwan e-mail: [email protected]

RAi ; RBi ; RCi ; RDi

RAj ; RBj ; RCj ; RDj

CCFC features located at the four corners of a rectangular multilayer PCB Circular centers of the CFCs at the ith layer Circular centers of the CFCs at the jth layer Nominal circular center distances between two of the CCFC features with the superscripts indicating related CCFC features Circular center distances between two of the CFCs at the ith layer with the superscripts indicating related CFCs Eccentric amounts of the CCFC features between the ith and jth layers with the superscripts indicating related CCFC feature Points on the moving platform Planar Cartesian coordinate system fixed on the CCD Planar Cartesian coordinate system fixed on the moving platform Positional vectors of the circular centers of the CFCs at the ith layer represented in the OmpXmpYmp coordinate system Positional vectors of the circular centers of the CFCs at the jth layer represented in

996

Δ+ Δ− Δ°

Int J Adv Manuf Technol (2010) 51:995–1008

the Omp-XmpYmp coordinate system Specified positive tolerance amount for the expansion defect Specified positive tolerance amount for the contraction defect Specified positive tolerance amount for the offset defect

1 Introduction Printed circuit boards (PCBs) are key components used within the electronics industry. Nowadays, elaborate PCBs have been extensively applied to high-added value electronic products such as personal computers, notebook computers, and mobile phones. In order to design and fabricate PCBs with complex and densely packed circuit layouts arranged in a limited space, multilayer PCBs have been developed and commonly adopted for over two decades owing to their high capacities. However, as the fabrication process for multilayer PCBs is still complicated, the inspection of fabricated multilayer PCBs is still considered an important task to help ensure quality control. As such, layer-to-layer misalignment phenomena of multilayer PCBs occurring during the laminate lay-up, bonding, and hot-pressing procedures [1, 2] are of great concern to PCB manufacturers. In such a lamination process, a stack of inner layers, or pre-fabricated copper-clad laminates (CCLs), is bonded together between prepregs (also called bonding sheets) under precisely controlled time, temperature, and pressure cycles using a laminating machine system. The prepregs are usually made of epoxy resinimpregnated fiberglass, with the epoxy resin being partially cured [3]. In the hot-pressing procedure, the CCL and prepreg layers must be heated and then bonded together with appropriate pressure levels being applied. Although alignment between layers is usually controlled with pins and registration holes, layer-to-layer misalignment phenomena cannot be eliminated completely. These phenomena are caused by several factors, such as the heating temperature and time control, the pressure control, the material properties of the epoxy resin, and the number of layers, which will influence the stability of heat transfer in the CCL and prepreg layers, in addition to the uniformity of the cured resin and the extent of residual stresses between adhesive layers. Three critical layer-to-layer misalignment modes, called expansion, contraction, and offset, occur during lamination possibly caused by these factors, and the laminated multilayer PCBs must therefore be inspected in order to check whether the required specifications are met. Furthermore, the misalignment inspection also improves the fabrication process. In other words, misalignment modes

must be evaluated in order to identify the optimal time, temperature, and pressure cycles for lamination. In practice, misalignment inspection of laminated multilayer PCBs cannot be undertaken using traditional destructive or optical inspection methods [4–6]. Instead, an X-ray imaging system provides a non-destructive and feasible means of inspecting the inner features and solder joints of PCBs [1, 7–12] as well as defects of ball grid arrays [13, 14]. Recently, Tick and Jantunen [15] evaluated the layer alignment and tape deformation in multilayer ceramic circuit boards using an X-ray inspection system. For layer-to-layer misalignment inspection, X-ray images containing information regarding the concentric copper foil circle (CCFC) features in certain multilayer PCBs have been used [8, 16]. In such cases, within each copper foil layer of a rectangular multilayer PCB, at least four copper foil circle (CFC) features are located at each of its four corners. An X-ray image of a CCFC feature at one of the four corners is formed, in which the CFC features in every copper foil layer are superimposed. Examining the concentricity of a CCFC feature from an X-ray image can assist in obtaining related information regarding local layer-to-layer misalignment of a multilayer PCB. Chen et al. [8] proposed an active contour algorithm that can effectively detect the circular information of CCFC features in an X-ray image. However, without measuring the center distances between CCFCs, the detected circular information itself can only indicate local conditions of a multilayer PCB. Such a process of the local condition evaluation achieves the so-called stationary CCFC inspection of PCBs. Nowadays, most commercially available X-ray systems used for PCB inspection can only provide the function of stationary CCFC inspection, and few can be applied to fully examine the expansion, contraction, and offset phenomena. In this study, an automated X-ray machine vision system was developed exclusively for the inspection of the layerto-layer misalignment of multilayer PCBs. An automated misalignment inspection process based on a mechatronics system and X-ray image processing techniques was also established. Experiments meant to inspect the expansion, contraction, and offset phenomena in ten-layer PCB samples were then conducted in order to test the feasibility of the developed inspection system.

2 Inspection system setup To nondestructively inspect the layer-to-layer misalignment of multilayer PCBs, an automated X-ray machine vision system was developed. Figure 1 shows the constructed machine inspection system, in which the main structure of the machine consisted of a welded tubular steel base frame combined with a granite surface plate. The granite surface plate was characterized by a rigid platform with a gantry-type

Int J Adv Manuf Technol (2010) 51:995–1008 Fig. 1 The inspection system constructed: (a) the solid CAD model of the naked machine and (b) the naked machine with the fixture device for printed circuit boards

Fig. 2 Functional block diagram of the inspection system

997

998

substructure. Based on the main structure, a three-dimensional motion system with a four-axis arrangement was designed. Four precision ball screw linear tables were mounted to the main structure in order to generate orthogonal linear motions. The x-axis linear table drove a moving platform to translate along the x-direction. The y1-axis linear table drove the z-axis linear table and an attached charge-coupled device (CCD) camera to translate along the y-direction. Meanwhile, the y2axis linear table drove an X-ray tube to translate along the ydirection. The y1- and y2-axis linear tables were parallel to each other. The z-axis linear table drove the CCD camera to translate along the z-direction for focusing adjustment. A fixture device for clamping inspected PCBs was then set on the moving platform. In Fig. 1, only the naked machine is shown for illustrative purpose. In practice, however, the whole machine was completely covered by lead shielding to prevent X-ray emission. Figure 2 shows the overall functional block diagram of the inspection system. The developed machine was functionally divided into a mechatronics system module and a machine vision module. Both modules were controlled by a host PC. In the mechatronics system module, a National Instruments (NI) PCI-7344 four-axis motion control card was installed in the host PC to manipulate four Oriental Motor five-phase step motors (a PK566AW motor for the x-axis motion; two PK569AW motors for the y1- and y2-axis motions, respectively; and a PK545AW motor for the z-axis motion). The step motors were driven through four Oriental Motor fivephase micro-step drivers (three RKD514L-C drivers for the x-, y1-, and y2-axis step motors, respectively, and an RKD507-A driver for the z-axis step motor). The step motors then directly drove the four ball screw linear tables, whose reference positions were measured by Renishaw RGH41X optical linear encoders and whose extreme positions were detected by limit switches. Through closedloop motion control, the machine had a resolution of 1 μm and was validated to have a kinematic accuracy of ±2 μm for Fig. 3 Diagram of the automated misalignment inspection process

Int J Adv Manuf Technol (2010) 51:995–1008

each axis. In the machine vision module, an NI IMAQ PCI1409 frame grabber was installed in the host PC to function as an image-grabbing card and capture the digital image sensed by a Hamamatsu H8481-05 CCD camera. Meanwhile, the X-ray beam source was generated by an Oxford Instruments XTF5011/75 X-ray tube. The captured digital image consisted of a 640×480 array of pixels with grayscale intensity values ranging from 0 to 255. By means of spatial calibration [17, 18], the conversion factor of the captured digital image corresponding to its real world dimension was found to be 6.29 μm/pixel. By employing the subpixel localization algorithm [18, 19], the developed X-ray image system could achieve an estimated resolution of 1/25 pixel [18], that is, 0.25 μm. To operate the inspection system through the host PC, human–machine interface software and image processing programs were developed and integrated in the NI LabView environment. For the purpose of inspection, the laminated multilayer PCB was placed on the fixture platform shown in Fig. 1(b) and then gently clamped by the fixture device. The x-, y1-, and y2-axis linear tables were then moved simultaneously, resulting in the PCB being moved to a location between the CCD camera and the X-ray tube. The lens axis of the CCD camera and the centerline of the X-ray outlet were maintained to be collinear as accurately as possible, i.e., the motions of the y1- and y2-axis linear tables were synchronized. After the inspected PCB reached the required location, the X-ray beam generated by the tube penetrated the PCB to produce a shadow of the CCFC feature, which was then sensed by the CCD camera. In this manner, X-ray images of the CCFC features located at the four corners of a multilayer PCB were sequentially captured. The captured X-ray images and the positional information of the x- and y1-axis linear tables measured by the optical linear encoders would be further employed to evaluate the layer-to-layer misalignment phenomena of the inspected PCB.

Int J Adv Manuf Technol (2010) 51:995–1008

999

3 The automated misalignment inspection process Based on the developed inspection system, an automated misalignment inspection process for multilayer PCBs was established and is described in this section. As shown in Fig. 3, considering that a multilayer PCB is placed on the moving platform and clamped by the fixture device for inspection, a planar Cartesian coordinate system, OmpXmpYmp, is set up on the moving platform. Another planar Cartesian coordinate system, OCCD-XCCDYCCD, is set up on the field of view (FOV) of the CCD camera with its origin at the upper-left corner of the FOV (note that the YCCD component is defined as positive downward for computer images). First, through simultaneous actuation of the x-, y1-, and y2-axis linear tables, the CCD camera and the moving platform are both moved so that the origin OCCD coincides with a point OA on the moving platform, upon which the upper-left CCFCs (labeled as A) of the PCB can appear in the FOV. Then, the digital image of the CCFCs A is captured, and by employing image processing algorithms, the positions of the circular centers of the CCFCs A are detected. For the ith layer, the positional vector of circular center Ai represented in the Omp-XmpYmp coordinate system can be calculated by:



R Ci

 x Ci ¼ Omp Ci ¼ ¼ Omp OC þ OC Ci yCi hmpi      xO C 1 0 x Ci ¼ þ yOC hmpi yCi hCCDi 0 1     xO C x Ci ¼ þ yOC hmpi yCi hmpi

ð3Þ



R Ai

 xA i ¼ Omp Ai ¼ ¼ Omp OA þ OA Ai yAi hmpi      xA i xO A 1 0 ¼ þ yOA hmpi yAi hCCDi 0 1     xAi xO A ¼ þ yAi hmpi yOA hmpi

using the same process. The positional vectors of the circular centers Bi, Ci, and Di, as represented in the Omp-XmpYmp coordinate system, denoted RBi , RCi , and RDi , respectively, can thus be calculated by:   x Bi RBi ¼ Omp Bi ¼ ¼ Omp OB þ OB Bi yBi hmpi      xO B 1 0 x Bi ¼ þ ð2Þ yBi hCCDi 0 1 yOB hmpi     xO B x Bi ¼ þ yOB hmpi yBi hmpi

ð1Þ

where the subscripts and refer to positional vectors represented in the Omp-XmpYmp coordinate system and the OCCD-XCCDYCCD coordinate system, respectively. The vector Omp OA ¼ fxOA yOA gThmpi, whose components xOA and yOA can be directly read from the optical linear encoders installed for the x- and y1-axis linear tables, represents the position of point OCCD in the Omp-XmpYmp coordinate system. Meanwhile, the vector OA Ai ¼ fxAi yAi gThmpi, whose components xAi and yAi are obtained from the captured X-ray image using image processing algorithms, represents the relative position between points Ai and OCCD in the Omp-XmpYmp coordinate system. Afterwards, the CCD camera and the moving platform are both moved so that the origin OCCD coincides with a point OB on the moving platform, upon which the upper-right CCFCs (labeled as B) of the PCB appears in the FOV to allow the capture of the digital image of the CCFCs B. The digital images of the bottom-right CCFCs (labeled as C) and the bottom-left CCFCs (labeled as D) of the PCB are also captured in turn

Fig. 4 Flowchart of the approach to locate the concentric circular centers from the concentric copper foil circle feature in an X-ray image

1000

Int J Adv Manuf Technol (2010) 51:995–1008



R Di

 xD i ¼ Omp OD þ OD Di yDi hmpi      xD i xO D 1 0 ¼ þ yOD hmpi yDi hCCDi 0 1     xD i xO D ¼ þ yDi hmpi yOD hmpi

After obtaining the positional vectors RAi , RBi , RCi , and RDi , the four center distances of the ith layer can be defined and calculated by:

¼ Omp Di ¼

Fig. 5 An illustrative example of the sequential procedure for detecting circular information from the concentric copper foil circle feature in an X-ray image

ð4Þ

ðABÞ

¼ k R Bi  R A i k

ð5Þ

ðBCÞ

¼ k R Ci  R Bi k

ð6Þ

di di

Int J Adv Manuf Technol (2010) 51:995–1008

1001

Fig. 6 X-ray images of the concentric copper foil circles (CCFCs) located at the four corners of an inspected ten-layer printed circuit board sample: (a) CCFCs A, (b) CCFCs B, (c) CCFCs C, and (d) CCFCs D

Table 1 Uncertainty test results

ðABÞ

d2

(mm)

ðBCÞ

d2

(mm)

ðCDÞ

d2

(mm)

ðDAÞ

d2

(mm)

Test 1 Test 2

494.787 494.769

355.360 355.323

494.806 494.785

355.342 355.316

Test 3 Test 4 Test 5 Test 6 Test 7 Test 8 Test 9 Test 10 Mean value Nominal value Systematic error Standard deviation Uncertainty

494.793 494.818 494.783 494.799 494.772 494.780 494.816 494.776 494.789 494.792 0.003 0.017 ±0.051

355.324 355.325 355.350 355.364 355.323 355.315 355.318 355.324 355.333 355.346 0.013 0.018 ±0.055

494.827 494.790 494.801 494.768 494.760 494.788 494.804 494.758 494.789 494.792 0.003 0.022 ±0.066

355.349 355.367 355.339 355.311 355.310 355.345 355.350 355.307 355.334 355.346 0.012 0.021 ±0.063

1002

Int J Adv Manuf Technol (2010) 51:995–1008

ðCDÞ

¼ k R D i  R Ci k

ð7Þ

ðDAÞ

¼ kR A i  R D i k

ð8Þ

di

di

ðABÞ

ðCDÞ

where di and di are referred to as the horizontal ðBCÞ center distances of the rectangular PCB, and di and ðDAÞ di the vertical ones. In addition, the four relatively eccentric amounts between the ith and the jth layers can be defined and calculated by:

ðAÞ eij ¼ RAj  RAi ð9Þ

ð BÞ eij ¼ RBj  RBi

ð10Þ

ð CÞ eij ¼ RCj  RCi

ð11Þ

ðDÞ eij ¼ RDj  RDi

ð12Þ

Step 2 Thresholding [18–21], which aims to separate objects (CCFCs) from the background: Based on the grayscale histogram of an X-ray image, the gray levels of the CCFCs and the background can be partitioned into two groups. By finding the wave trough located between the two groups, a proper threshold value is then selected to extract the CCFCs from the background. The thresholded image is a transformed binary image with the objects in pure white color (whose pixel intensity

Using the information obtained from the calculated center distances and relatively eccentric amounts from every copper foil layer, the layer-to-layer misalignment phenomena can be evaluated. Such a misalignment inspection process can be programmed and automatically repeated using the developed inspection system for the examination of other PCBs with identical specifications. The image processing procedure and the misalignment mode analysis used for the established misalignment inspection process are described below. 3.1 Image processing procedure The image processing procedure used to detect the positions of the circular centers of the CCFCs from the captured X-ray images is introduced here. Based on the NI LabView/Vision Assistant environment, the image processing programs were developed. Figure 4 shows the flowchart of the approach used to locate the concentric circular centers from the CCFC feature in an X-ray image. Once an original X-ray image has been captured and read, the sequential image processing procedure described below is executed. Step 1 Image filtering, which aims to reduce random impulse noise, also called salt and pepper noise [19–21], appearing in the original X-ray image: In multilayer PCB inspection, salt noise frequently appears in the original X-ray images and can be reduced by using order-statistics filters [20, 21]. In this step, the median filter is applied to reduce salt noise while simultaneously retaining the sharpness of edges.

Fig. 7 Inspection results of an expansion case: (a) horizontal center distances, (b) vertical center distances, and (c) relatively eccentric amounts

Int J Adv Manuf Technol (2010) 51:995–1008

1003

value is 1) and the background in pure black color (whose pixel intensity value is 0). Step 3 Morphological operation, which aims to modify and reshape the boundaries of objects in the transformed binary image: In this step, the opening operation [18–21] is employed to smoothen the contours of objects, as well as break narrow isthmuses and eliminate thin protrusions and isolated particles. The narrow isthmus, thin protrusion features, and isolated particles usually occur after performing thresholding. Step 4 Two-dimensional edge detection [17, 18, 22], which aims to locate the boundaries and circular centers of the CCFCs: A circular edge detection function within NI Vision Assistant, the spoke function [18], is applied to find the circular centers of the CCFCs. By assigning an annular search area (also called a region of interest) covering the CCFCs, radial search lines with equal angular intervals will outwardly detect circular edges by finding the first location of a sudden change in pixel intensity value along the direction of each search line. Then, by applying the least square circle fitting method [17, 19] with all detected edge points being involved, the radius and circular center of a CFC can be obtained. In this step, due to the limitation of the spoke function, only the innermost circular feature of the CCFCs will be detected by executing the spoke function once. Step 5 Image masking with morphological operation, which aims to eliminate the region occupied by the detected CFC from the binary image: After the innermost circular feature has been detected in Step 4, the area within the fitted circle is used as a mask (a logical operator [20, 21]) to turn all pixel intensity values (with a value of 1) within the circular area into the background-level values (with a value of 0). In Table 2 Inspection data of an expansion case

Center distance (in average) Layer Layer Layer Layer

2 4 6 8

(i=2) (i=4) (i=6) (i=8)

essence, this can be regarded as an image subtraction operation [20] for the circular area itself. The opening operation is then employed to remove residual isolated particles that occur after performing image masking. Step 6 Image negating [20], which aims to reverse the pixel intensity values of the binary image after masking: This operation is necessary for sequential image masking operations. Step 7 Check if all circular centers have been located. If not, repeat Steps 4 to 6 until all CFC features are detected and eliminated. Finally, save and print the positions of all detected circular centers. Figure 5 shows an illustrative example of the abovementioned procedure being applied to detect the circular centers of the CCFCs from an X-ray image. In Fig. 5(a), two CCFCs can be observed in the original image, the inner a solid CFC, the outer an annular one. The solid CFC has a single circular boundary that can be detected, while the annular CFC has two boundaries (the inner and outer circles of the annulus) that can be detected. For the annular CFC, the representative circular center position will be evaluated by averaging the detected inner and outer circular center positions. The images obtained after performing filtering and thresholding are shown in Fig. 5(b) and (c), respectively, and that obtained after performing the opening operation is shown in Fig. 5(d). Figure 5(e) shows the inner CFC being detected through the use of the spoke function, in which the edge points are detected to locate the circular center using the least square circle fitting method. Then, the masked and negated binary images are shown in Fig. 5(f) and (g), respectively; as can be seen, the inner CFC has been eliminated. Figure 5(h), (i), and (j) sequentially show the procedure of detecting the inner circle of the annular CFC and the subsequent steps of masking and negating the image. Finally, Fig. 5(k) and (l) similarly show the procedure of detecting the outer circle of the annular

ðABÞ

di (mm) 494.789 494.789 494.781 494.879

Relatively eccentric amount (in average) ðAÞ eij (mm) Layer 2 (i=2, j=2) 0 Layer 4 (i=2, j=4) 0.025 Layer 6 (i=2, j=6) 0.022 Layer 8 (i=2, j=8) 0.056

ðBCÞ

di (mm) 355.321 355.307 355.316 355.330

ðBÞ

eij (mm) 0 0.017 0.022 0.039

ðCDÞ

di (mm) 494.804 494.762 494.784 494.886

ðCÞ

eij (mm) 0 0.014 0.017 0.053

ðDAÞ

di (mm) 355.290 355.317 355.313 355.310

ðDÞ

eij (mm) 0 0.029 0.011 0.036

1004

Int J Adv Manuf Technol (2010) 51:995–1008

CFC, followed by masking the image. As all CFC features have been detected and eliminated, as shown in Fig. 5(l), the image processing procedure is considered complete, and the detected circular center positions can be further used for inspection purposes. In Step 3 and Step 5 of the proposed image processing procedure, the major purpose of using the opening operation is to eliminate isolated particles distributed between the CFC features. Such isolated particles may cause mistakes when the circular edge detection function described in Step 4 is employed. If the isolated particles are not eliminated, erroneous points may be detected using the spoke function as shown in Fig. 5(e), (h), and (k). Hence, the use of the opening operation is quite necessary for the image processing procedure. The other purpose of using the opening operation is to eliminate the narrow isthmus and thin protrusion features occurring at the boundaries of the CFCs. Although the reshaped boundaries of the CFCs will deviate slightly from the original ones in some parts, this operation should render the reshaped boundaries smoother and closer to their actual geometric shapes. In addition, when the number of sampling points used for the circular edge detection is sufficient, i.e., when the angular interval between the radial search lines is small enough, the effect of the relative deviations between the original and reshaped boundaries of the CFCs on influencing the circular center locations is reasonably reduced and could be considered negligible.

ðCDÞ

  > d ðCDÞ þ Δþ

ð15Þ

ðDAÞ

  > d ðDAÞ þ Δþ

ð16Þ

di

di

where d(AB), d(BC), d(CD), and d(DA) are the corresponding nominal center distances, and Δ+ is the specified positive tolerance amount for the expansion defect.

3.2 Misalignment mode analysis As mentioned in the introduction, three layer-to-layer misalignment modes of multilayer PCBs, referred to as expansion, contraction, and offset, are considered important issues by PCB manufacturers. Based on the calculated center distances and relatively eccentric amounts, the misalignment modes of inspected PCBs can be analyzed. In practice, for a qualified multilayer PCB, the deviations of the actual center distances and relatively eccentric amounts must lie within their specified tolerance bands as compared with their nominal (designed) ones. Such a requirement also helps to establish the criteria for the analysis of misalignment modes. The expansion defect occurs when one of the actual center distances is larger than its upper limit (the nominal distance plus the specified tolerance amount). Considering the ith layer of a PCB being inspected, an expansion defect would occur if at least one of the following four inequalities is satisfied:   ðABÞ di > d ðABÞ þ Δþ ð13Þ

ðBCÞ

di

  > d ðBCÞ þ Δþ

ð14Þ

Fig. 8 Inspection results of a contraction case: (a) horizontal center distances, (b) vertical center distances, and (c) relatively eccentric amounts

Int J Adv Manuf Technol (2010) 51:995–1008

1005

Likewise, the contraction defect occurs when one of the actual center distances is smaller than its lower limit (the nominal distance minus the specified tolerance amount). Considering the ith layer of a PCB being inspected, a contraction defect would occur if at least one of the following four inequalities is satisfied: ðABÞ

  < d ðABÞ  Δ

ð17Þ

ðBCÞ

  < d ðBCÞ  Δ

ð18Þ

ðCDÞ

  < d ðCDÞ  Δ

ð19Þ

di

di

di

ðDAÞ

di

  < d ðDAÞ  Δ

ð20Þ

where Δ− is the specified positive tolerance amount for the contraction defect. Finally, the offset defect occurs when one of the relatively eccentric amounts is larger than the specified tolerance amount. Considering the ith layer of a PCB as the datum layer and the offset of the jth layer of the same PCB relative to the datum layer being inspected, an offset defect would occur if at least one of the following four inequalities is satisfied: ðAÞ

ð21Þ

ð BÞ

ð22Þ

eij > Δ eij > Δ

Table 3 Inspection data of a contraction case

Center distance (in average) Layer Layer Layer Layer

2 4 6 8

(i=2) (i=4) (i=6) (i=8)

ð CÞ

ð23Þ

ðDÞ

ð24Þ

eij > Δ

eij > Δ

where Δ° is the specified positive tolerance amount for the offset defect. In a multilayer PCB, the three misalignment modes may exist at the same time, presenting a composite situation that must be evaluated by simultaneously considering all inequalities represented by Eqs. 13–24.

4 Results and discussion To test the feasibility of the developed X-ray machine vision system combined with the automated misalignment inspection process, experiments meant to inspect multilayer PCBs were conducted. In this study, samples of ten-layer PCBs that are 515×375 square millimeters in size were examined. The top and bottom layers of the samples were called the first and tenth layers, respectively. Each ten-layer PCB had four copper foil layers, that is, CCFC features were printed on the second, fourth, sixth, and eighth layers. Figure 6 shows the original X-ray images (partly cropped) of the CCFCs located at the four corners of an inspected ten-layer PCB sample. As can be seen, two crossed lines intersecting at the center of the FOV are additionally marked as positional references. The nominal center distances were given as d(AB) =d(CD) = 494.792 mm (19,480 mil) and d(BC) =d(DA) =355.346 mm (13,990 mil). The specified tolerances, Δ+, Δ−, and Δ°, were all given as 0.076 mm (3 mil), that is, the specified tolerance band was ±0.076 mm (or ±3 mil). The experimental results and discussion are provided in the subsequent subsections.

ðABÞ

di (mm) 494.788 494.782 494.709 494.786

Relatively eccentric amount (in average) ðAÞ eij (mm) Layer 2 (i=2, j=2) 0 Layer 4 (i=2, j=4) 0.026 Layer 6 (i=2, j=6) 0.038 Layer 8 (i=2, j=8) 0.030

ðBCÞ

di (mm) 355.329 355.303 355.317 355.345

ðBÞ

eij (mm) 0 0.021 0.055 0.007

ðCDÞ

di (mm) 494.803 494.760 494.707 494.795

ðCÞ

eij (mm) 0 0.018 0.056 0.014

ðDAÞ

di (mm) 355.289 355.316 355.316 355.323

ðDÞ

eij (mm) 0 0.029 0.042 0.005

1006

4.1 Uncertainty tests Prior to the inspection of the PCB samples, a qualified tenlayer PCB template was inspected using the developed X-ray machine vision system in order to test its uncertainty or repeatability. X-ray images of the four CFCs on the second layer of the PCB template were repeatedly captured ten times in order to calculate the center distances and their uncertainties. The test results are listed in Table 1, in which the four ðABÞ ðBCÞ calculated center distances are represented by d2 , d2 , ðCDÞ ðDAÞ d2 , and d2 . As shown in the table, through the threestandard-deviation-band approach, the inspection system could achieve an uncertainty of ±0.066 mm (±2.598 mil). Although the uncertainty was merely slightly less than the specified tolerance band (±0.076 mm), the mean center distances were quite close to the nominal ones since their differences (i.e., the systematic errors) were within 0.013 mm. In other words, repeated inspection of a multilayer PCB and subsequent adoption of the mean values can provide an acceptable method of the inspection task. The major cause of the uncertainty of the inspection system should be the instability of the X-ray images. This arises because the quality of an X-ray image is strongly influenced by random impulse noise produced by the X-ray passing through the multilayer PCB consisting of CCLs and prepregs. The image filtering, thresholding, and morphological operation applied to reduce impulse noise and subsequent enhancement and smoothing of object features in X-ray images may also lead to a slight loss of information for the circular center detection, which causes unavoidable uncertainty.

Int J Adv Manuf Technol (2010) 51:995–1008

PCB, as all relatively eccentric amounts relative to the datum layer were less than 0.076 mm. This means that the inspected PCB contained horizontal expansion defects in the eighth layer, while the other three layers were qualified. 4.2.2 Contraction case Figure 8 shows the inspection results of a PCB sample that contained contraction defects, and the inspection data for which are also listed in Table 3. In this case, with the ðABÞ ðCDÞ exception of d6 (494.709 mm) and d6 (494.707 mm), which were smaller than their specified lower bound

4.2 Inspection of PCB samples Several PCB samples were inspected, and their misalignment modes were evaluated. In order to reduce the uncertainty in the inspection task, each sample was repeatedly inspected ten times, after which, the average data were taken for subsequent analysis. Three representative cases, in which the expansion, contraction, and offset phenomena were found, respectively, are described below. 4.2.1 Expansion case Figure 7 shows the inspection results of a PCB sample that contained expansion defects, and the inspection data for which are also listed in Table 2. As can be seen, with the ðABÞ ðCDÞ exception of d8 (494.879 mm) and d8 (494.886 mm), which were larger than their specified upper bound (494.868 mm), the other center distances all lay within their specified tolerance bands (494.792±0.076 or 355.346 ±0.076 mm). In addition, considering the second layer as the datum layer, no offset defect existed in the inspected

Fig. 9 Inspection results of an offset case: (a) horizontal center distances, (b) vertical center distances, and (c) relatively eccentric amounts

Int J Adv Manuf Technol (2010) 51:995–1008

1007

(494.716 mm), the other center distances all lay within their specified tolerance bands. In addition, considering the second layer as the datum layer, no offset defect existed in the inspected PCB, as all relatively eccentric amounts relative to the datum layer were less than 0.076 mm. That is, the inspected PCB contained horizontal contraction defects in the sixth layer, while the other three layers were qualified. 4.2.3 Offset case Figure 9 shows the inspection results of a PCB sample that contained offset defects, and the inspection data for which are also listed in Table 4. It was found that all center distances lay within their specified tolerance bands. However, considering the second layer as the datum layer, ðAÞ ð BÞ ð CÞ ðDÞ e28 (0.095 mm), e28 (0.086 mm), e28 (0.082 mm), and e28 (0.083 mm) were evidently larger than their specified upper bound (0.076 mm), while other relatively eccentric amounts relative to the datum layer were quite less than 0.076 mm. In other words, the inspected PCB contained offset defects in the eighth layer, while the other three layers were qualified. As evidenced by the above-described three cases, the expansion, contraction, and offset phenomena in the ten-layer PCB samples were successfully examined. The developed Xray machine vision system combined with the automated misalignment inspection process was validated for the misalignment inspection for certain multilayer PCBs.

5 Discussion The proposed misalignment inspection method of multilayer PCBs is in essence based on a combination of the circle detection procedure for the CCFCs in X-ray images and the positional information of the multi-axis motion system read from the installed optical linear encoders. The feasibility of Table 4 Inspection data of an offset case

Center distance (in average) Layer Layer Layer Layer

2 4 6 8

(i=2) (i=4) (i=6) (i=8)

such an inspection method was verified through experiments. Another possible inspection method based on the socalled feature-based image alignment or image mosaics should also be discussed. Image mosaic techniques have been developed by many researchers [23–29] for the automatic alignment of a collection of digital images with partial views into a broader and complete view. The image mosaic process is advantageous in terms of overcoming the limitation of the FOV of a camera to capture complete images of large objects. For the misalignment inspection of multilayer PCBs in which an image mosaicing algorithm is applied, a collection of X-ray images with partial views of an inspected PCB must be captured and then combined into a complete view, i.e., a panoramic X-ray image mosaic of the inspected PCB. The constructed image mosaic is then used to detect the positional information of the CCFCs in order to evaluate the misalignment phenomena of the inspected PCB. If image mosaicing algorithms are employed for the misalignment inspection, the positional information read from optical linear encoders is no longer involved. In other words, the misalignment inspection becomes a purely image processing-related task. Nevertheless, in order to practice image mosaicing for the misalignment inspection, three critical concerns must be considered. First, random impulse noise in X-ray images may influence the stability of feature point detection or geometric transformation for mosaicing. That is, if several collections of X-ray images of identical PCB sample are captured, a lower repeatability of its image mosaics may be unavoidable. Second, the pixel size of a mosaic image, which is crucial for the center distance calculation, may slightly lose its accuracy to reflect the real world dimensions of the inspected PCB. The use of positional information read from optical linear encoders should be more reliable than the use of the pixel size of a mosaic image. Third, performing image mosaicing usually involves much iterative computation and is therefore more timeconsuming than the proposed method. Such a disadvantage

ðABÞ

di (mm) 494.796 494.793 494.788 494.800

Relatively eccentric amount (in average) ðAÞ eij (mm) Layer 2 (i=2, j=2) 0 Layer 4 (i=2, j=4) 0.009 Layer 6 (i=2, j=6) 0.023 Layer 8 (i=2, j=8) 0.095

ðBCÞ

di (mm) 355.320 355.323 355.335 355.324

ðBÞ

eij (mm) 0 0.012 0.009 0.086

ðCDÞ

di (mm) 494.788 494.778 494.798 494.809

ðCÞ

eij (mm) 0 0.007 0.019 0.082

ðDAÞ

di (mm) 355.331 355.325 355.349 355.342

ðDÞ

eij (mm) 0 0.008 0.007 0.083

1008

may decrease the efficiency of the misalignment inspection process during mass production of multilayer PCBs. Therefore, evaluation of the industrial feasibility of applying image mosaicing for the misalignment inspection with respect to the above concerns could be meaningful future work.

Int J Adv Manuf Technol (2010) 51:995–1008

9.

10.

11.

6 Conclusions To inspect the layer-to-layer misalignment of laminated multilayer PCBs, an automated X-ray machine vision system was developed and constructed. Based on a mechatronics system and X-ray image processing techniques, an automated misalignment inspection process was then established. A procedure for effectively detecting the circular center locations of the CCFC features from the captured X-ray images as well as the criteria for the misalignment mode analysis have been introduced. Experiments meant to inspect the expansion, contraction, and offset phenomena in ten-layer PCB samples were also conducted in order to test the feasibility of the developed machine inspection system. From the experimental results, the developed X-ray machine vision system, in combination with the automated misalignment inspection process, was validated to be an applicable means of the misalignment inspection for certain multilayer PCBs.

12.

13.

14.

15.

16. 17. 18.

19. Acknowledgments The authors are grateful to the National Science Council of Taiwan for supporting this research under Grant No. NSC95-2221-E-011-211 and Grant No. NSC-96-2221-E-011-129. Technical advising and support of Yayatech Corp. for developing the X-ray inspection system is also remembered with gratitude.

20.

21.

References 1. Scarlett JA (ed) (1985) The multilayer printed circuit board handbook. Electrochemical Publications, Ayr, Scotland, pp 181– 235; pp. 423–429 2. Coombs CF Jr (1988) Printed circuits handbook—third edition. McGraw-Hill, New York, pp 31.1–34.18 3. DeGarmo EP, Black JT, Kohser RA (2003) Materials and processes in manufacturing—ninth edition (update edition). Wiley, Hoboken, New Jersey, pp 910–915 4. Wu WY, Wang MJJ, Liu CM (1996) Automated inspection of printed circuit boards through machine vision. Comput Ind 28(2):103–111 5. Rau H, Wu CH (2005) Automatic optical inspection for detecting defects on printed circuit board inner layers. Int J Adv Manuf Technol 25(9–10):940–946 6. Rankov V, Crispin AJ (2007) Automated inspection of PCB components using a genetic algorithm template-matching approach. Int J Adv Manuf Technol 35(3–4):293–300 7. Moganti M, Ercal F (1995) Automatic PCB inspection system. IEEE Potentials 14(3):6–10 8. Chen YS, Wu CH, Hu WF (2006) An active contour algorithm for detecting the circular features in a PCB x-ray image. In Proceedings

22.

23.

24. 25.

26. 27.

28.

29.

of the SPIE-IS&T 18th Annual Symposium of Electronic Imaging Science and Technology, Vol. 6070, pp. 224–231 Hanke RF, Weiss T, Petsch N (1992) Automated 3D X-ray inspection of fine pitch PCB’s. In Proceedings of the 13th IEEE/ CHMT International Electronics Manufacturing Technology Symposium, pp 187-190 Neubauer C (1997) Intelligent X-ray inspection for quality control of solder joints. IEEE Trans Compon, Packag, Manuf Technol Part C 20(2):111–120 Lehmann DK (2002) X-ray system for optimizing PCB inspection. Circuits Assembly 13(2):35–39 Manjeshwar PK, Craik J, Phadnis S, Srihari K (2006) Effectiveness study of an automated 3D laminography x-ray inspection system in a high-volume-low-mix SMT line. Int J Adv Manuf Technol 30(11–12):1191–1201 Rooks SM, Benhabib B, Smith KC (1995) Development of an inspection process for ball-grid-array technology using scannedbeam X-ray laminography. IEEE Trans Compon Packag Manuf Technol Part A 18(4):851–861 Sumimoto T, Maruyama T, Azuma Y, Goto S, Mondou M, Furukawa N, Okada S (2003) Development of image analysis for detection of defects of BGA by using X-ray images. In Proceedings of the 20th IEEE Instrumentation and Measurement Technology Conference, Vol. 2, pp. 1131–1136 Tick T, Jantunen H (2008) An X-ray imaging based layer alignment and tape deformation inspection system for multilayer ceramic circuit boards. IEEE Trans Electron Packag Manuf 31(2):168–173 Yayatech Corp., X-ray checker XC-25P. http://www.yayatech. com/index-en.asp Su JC, Tarng YS (2006) Measuring wear of the grinding wheel using machine vision. Int J Adv Manuf Technol 30(1):50–60 National Instrument Corp. (2007) NI vision concepts manual. National Instrument Corp., Austin, Texas, pp. 3.1–3.18; pp. 8.1– 8.15; pp. 9.1–9.31; pp 11.1–11.22 Jain R, Kasturi R, Schunck B (1995) Machine vision. McGrawHill, New York, pp. 25–72; pp. 112–185; pp. 210–223; pp. 309– 364 Gonzalez RC, Woods RE (2002) Digital image processing— second edition. Prentice-Hall, Upper Saddle River, New Jersey, pp. 75–146; pp. 519–566; pp. 595–611 Umbaugh SE (2005) Computer imaging: digital image analysis and processing. CRC Press, Boca Raton, Florida, pp 67–199; pp 341–480 Su JC, Tarng YS (2008) Automated visual inspection for surface appearance defects of varistors using an adaptive neuro-fuzzy inference system. Int J Adv Manuf Technol 35(7–8):789–802 Szeliski R (1994) Image mosaicing for tele-reality applications. In Proceedings of the 2nd IEEE Workshop on Applications of Computer Vision, pp 44-53 Szeliski R (1996) Video mosaics for virtual environments. IEEE Comput Appl 16(2):22–30 Bao P, Xu D (1999) Complex wavelet-based image mosaics using edge-preserving visual perception modeling. Comput Graph 23 (3):309–321 Zhao L, Yang YH (1999) Mosaic image method: a local and global method. Pattern Recognit 32(8):1421–1433 Can A, Steward CV, Roysam B, Tanenbaum HL (2002) A featurebased technique for joint, linear estimation of high-order imageto-mosaic transformations: mosaicing the curved human retina. IEEE Trans Pattern Anal Mach Intell 24(3):412–419 Tian GY, Gledhill D, Taylor D (2003) Comprehensive interest points based imaging mosaic. Pattern Recognit Lett 24(9–10): 1171–1179 Kim DH, Yoon YI, Choi JS (2003) An efficient method to build panoramic image mosaics. Pattern Recognit Lett 24(14):2421–2429