IRJET- Automatic Identification, Analysis and Investigation of Printed Circuit Boards for Defects and Errors Disclosure and Classification based on Nature

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 12 | Dec 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 409
AUTOMATIC IDENTIFICATION, ANALYSIS AND INVESTIGATION OF
PRINTED CIRCUIT BOARDS FOR DEFECTS AND ERRORS DISCLOSURE
AND CLASSIFICATION BASED ON NATURE
Mithilesh Padhen1
1Student, Dept. of Electronics and Telecommunication Engineering, Pimpri Chinchwad College of Engineering,
Pune, Maharashtra, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract –In the era of electronics, computers and high -
end machine and various high-performance devices the
printed circuit boards and integral part of any equipment.
This printed circuit boards plays a very important role in
smooth functioning of any devices. So, for successful
operation of any equipment this printed circuit boards
should be properly tested, inspected and investigated. Also,
Inspection and Investigation of printed circuit board (PCB)
has been a crucial process in the electronic manufacturing
industry to guarantee product quality & reliability, cut
manufacturing cost and to increase production. The PCB
inspection involves detection of defects and errors in the
PCB and classification of those defects and errors in order
to identify the roots of defects. In this paper, all 14 types of
defects are detected and are classified in all possibleclasses
using referential inspection and investigation approach.
The proposed algorithm is mainly divided into five stages:
Image registration, Pre- processing, Image segmentation,
Defect detection and Defect classification. The proposed
algorithm is able to perform inspection even when the
various operations are done on the test image. The various
operation on captured test image is rotated, scaled and
translated withrespect to template image which makesthe
algorithm rotation, scale and translation in-variant. The
novelty of the proposed algorithm lies in its robustness,
reliability and efficiency to analyze a defect in its different
possible appearance and severity. In addition to this,
algorithm takes only 2.528 s to inspect and investigate a
PCB image. The efficiency and reliability of the proposed
algorithm is verified by conducting experiments on the
different PCB images and it shows that the proposed
algorithm is suitable for automatic identification visual
inspection of PCBs.
Key Words: Printed Circuit Boards,Automatic Visual
Inspection, Detection and Inspection, Machine Vision
and Classification.
I. INTRODUCTION
Production and manufacturing of Printed Circuit Boards
is an essential component in the electronics and
semiconductor industries. The performance and
efficiency of a PCB is significantlydependentonitsquality
and reliability. A defective PCB may result in undesirable
circuit behavior and may end upin a defective, unwanted
and unreliable product. Due to this Printed Circuit Board
inspection and investigation is a crucial process in
electronics industries. The aim of this inspection process
is to assure 100% quality and reliabilityof all parts, which
costs the most in manufacturing [1], [2]. Conventionally,
human operators are involved in the visual inspection of
PCBtodetect and classify the defects and various types of
errors occurring and unwanted noises. This conventional
manual inspection and investigation process is time-
consuming, tedious and error-prone. Also, the results of
inspection and investigation may vary person to person
due to human inconsistency and operating nature. The
quality control problem can be solved by using
developments in advancedcomputer vision field. In order
to make PCB inspection and investigation process fast,
reliable and efficient, automatic visual inspection (AVI)
systems is more useful in various types of electronics
industries.
Automatic Visual Inspection (AVI) based approaches are
mainly divided into three different categories: The first is
referential, second is non-referential and last one is
hybrid methods [3]. Considering the first case of
referential method, the given test image of the PCB is
compared with its predefined template image in order to
locate and finding out various defects. In another case of
non-referential method which is basedon thedesignrule-
based method which verifies whether the design ofPCB is
in predefined limits or not. But the disadvantage of the
non-referential method is that it is not able to identify
defects in their distorted appearance. The hybrid method
is most advanced one. The hybrid method is generallythe
combination of both referential and non-referential
methods. But, the disadvantage of thehybridmethodisits
higher and advanced computational complexity. The
sample template and sample defective images of PCB are
shown in Fig. 1(a) and (b), respectively. There are 14
types of various underlying known defects in PCB as
shown in Fig. 1(b).

(a) Template image of PCB
(b) Test image of PCB with defects:
Fig. 1: PCB images for referential method
(1) Breakout, (2) Pinhole, (3) Open circuit, (4) Under
etch, (5) Mouse bite, (6) Missing conductor, (7) Spur,
(8) Short, (9) Wrong size hole, (10) Conductor too
close, (11) Spurious Copper, (12) Excessive short,
(13) Missing hole and (14) Over etch
In the literature survey, numerous authors tried to
discloseand classify the major possible occurring defects
in generated PCB image using different methods. Wuetal.
[4] used the referential method in order to disclose and
classify the defects into various types of seven defined
groups. The classification is performed according to three
indices of a defect based on type and number of objects.
Putera et al. [5] utilize the area property of defect for
classifying it into seven defined groups, with maximum
allowable four defects in a group. Further,Nakagawa et al.
[6] propose a differential method and it classifies the
defects into three defined classes. The research
articulated in [6] differentiates the PCV image with the
help of multiple support vector machine (SVM) which is
trained with 24 various features of defect candidate. In
[7], authors propose a referential method by using the
edge grey gradient of the PCB image in order to classify
defects into 5 defined classes. Furthermore, Kumar et al.
[8] propose a non-referential method for further
classification of defects into 4 defined classes. While, this
method is having disadvantage that it can classify only one
defect per image. The classificationof defects in their class
is as crucial as detection ofdefects. This classification is a
naturalized process in order toidentifytherootsandbasics
of defects. As per the literaturesurveynoauthor has tried to
classify all various types of 14 PCB defects into all 14
possible classes.
Fig. 2: Block schematic of the proposed algorithm
In this paper, I propose a referential method to disclose
and classify the occurring defects of PCB into all possible
14 classes. Theproposed algorithm is mainly divided into
five operationalstages: Imageregistration, Pre-processing,
Image segmentation, Defect detection and Defect
classification. Firstly, in Section II, image registration
technique is articulated in order to remove unnecessary
variation in captured test image like rotation, angular
position, scale and translation with respect to template
image of the same PCB. Next to that in Section III, pre-
processing steps are elaborated in order to reduce noise
factor, increase the efficiency and enhance the image
quality. In Section IV,theimage segmentationis produced.
The defect detection and classification are the topics of
discussion in Sections V and VI, respectively. At last,
observed results and generated timing report of the

algorithm is shown in Section VII. Finally, conclusion is
given in Section VIII. The complete block schematic of the
proposed algorithm is shown in Fig. 2.
Fig. 3: Un-registered test PCB image
Fig. 4: Block diagram of image registration
process
II. IMAGE REGISTRATION
Sample Test PCB is scanned by HP LaserJet scanner in
order to generate the test PCB image. This image may
have variations to an extent in terms of rotation, angular
position and translation with respect to the template
image as shown in Fig. 3. Such variations can be abolished
by using image registration techniques [9]. The proposed
block diagram of image registration process is shown in
Fig. 4. The generated test image and template images are
converted into grey scale by with the help of Eq. (1)
Greylevel = 0.299 · R + 0.587 · tt + 0.114·B, (1)
TABLE I
REGISTRATION TIME USING DIFFERENT
FEATURE EXTRACTION METHODS
Feature Extraction Method Execution Time (s)
SURF [11] 2.04
Harris [12] 2.635
BRISK [13] 1.497
FAST [10] 1.143
MSER [14] 4.411
MinEigen [15] 5.2
Fig. 5: Output of image registration process
where R, tt and B are the red, green and blue channels in
color image. Next to this is the process of extraction of the
features from the both present template and test images.
Since this process is mosttime-consumingandlonglastingin
nature in image registration algorithm it is desirable to use
high speed computational algorithm for this. Table I shows

required time to execute registrationprocessusingdifferent
feature extraction techniques and methods. Features from
accelerated segment test (FAST)algorithm [10] is usedsince
it takes lowest time comparing to other present extraction
methods as shown in Table I. The extracted features are
matched and verified using sum of squared difference (SSD)
metric. Geometric transformation matrix is then estimated
from matched features using m-estimatorsampleconsensus
(MSAC) algorithm [16].Theestimatedtransformationis then
enforced to test image in order to generate the registered
image. The output of image registration is depicted in Fig.
5.
III. PRE-PROCESSING
The acquired PCB images may have presence of noises
such assalt andpeppernoise.Also, theseimagesmayhave
high variations in intensity levels due to different lighting
position and brightening exposure, whicheventuallyleads
to improper binarization of image. The objective of pre-
processing is to remove noise and enhance the image
details and improve the efficiency. Fig. 6(a) depicts the
grey scale image of PCB using Eq. (1). Median ﬁlter of
mask size 7*7 is then enforced on to the grey scale image
for the purpose of removing salt and pepper noise. The
output image is shown in Fig. 6(b). Next to the process of
removal of noise, high-intensity variation is suppressed
to an extent by applying Gaussian low pass filtering
method having standard deviation = 1. In Fig. 6(c) we
have shown a gaussian low pass filtered image.
(a) Test image in grey scale
(b) Output of median filtering
(c) Output of low pass filtering
Fig. 6: Preprocessing steps
IV. IMAGE SEGMENTATION
Succeeding to pre-processing step, there is occurrence of
image segmentation. The objective of image segmentation is
to exemplify the image in different parts(asinsetsofpixels),
which makes the representation of image more substantial.
In PCB image, there are mainly three important parts: (1)
wiring tracks (2) soldering pads and (3) holes. In the
proposed approach, we use approach of histogram
thresholding method,followedbymathematical morphology
operations to divide the PCB image into mentioned parts.
Fig. 7 shows the normalized histogram of the PCB image.
Wiring tracks and soldering pads are produced by using
upperand lower threshold points asshowninEx.(2)and(3),
resp.
Wiring tracks = -1, if 95 < greylevel < 140; (2)
0, else
(b) Output of median filtering
Fig. 7: Normalized histogram of PCB image

(a) Wiring tracks (b) Soldering pads
(c) Holes
Fig. 8: Segmented images
Soldering pads =-1, if greylevel >140; (3)
0, else
The zero regions inside the soldering pads show holes.
These zero regions are now full of with region filling
operations. Soldering pads regions are then subtracted
from this filled image in order to produce the regions of
holes. The segmented images are shown in Fig. 8.
V. DEFECT DETECTION
The segmented images (including wiring tracks,soldering
pads and holes) of test and template images have
difference in each other due to defects occurring in testing
PCB image. So, the defects can be simply disclosed by
process of image subtraction. These defects generally are
of two types: (1) positive defects (PD) and (2) negative
defects (ND). As shown in Eq (4) positive defects can be
disclosed by subtracting segmentedtemplateimagesfrom
the corresponding segmented testing images; and vice
versa for negative defects Eq (5)
PDi = testingi –templatei (4)
NDi = templatei –testingi (5)
where, i gives idea of wiring, tracks, soldering, pads and
holes. Uneven binarization of edges also produces small
differences between test and template images. This kind
of small differences can be removed by method of area
filtering. Discloseddefectsafterareafiltering aredepicted
in Fig. 9.
(a) PD- Wiring tracks (b) ND- Wiring tracks
(c) PD- Soldering pads (d) ND- Soldering pads
(e) PD- Holes (f) ND- Holes
Fig. 9: DefectDetection
TABLE II
DEFECTS RELATED TO WIRING TRACKS
Positive Defects (PWD) Spur, Short, Spurious
copper, Excessive
short, Conductor too
close
Negative Defects (NWD) Pinhole, Mouse bite,
Open circuit, Missing
conductor, Conductor
too close

VI. DEFECT CLASSIFICATION
A. Defects Related to Wiring Tracks
Positive and negative defects related to wiring tracks are
shown in Table II. Centroid and maximum radius of
defects
Fig.10: Classification of wiring track defects (positive)
are obtained from PDW and NDW images by adopting 8-
connected components. Toverify the neighborhood of a
defect, a square region (where length = maximumradius
of defect, center =centroid of defect) is croppedfromthe
divided wiring track image of template image (WT). The
flowchart of defect classification is depicted in Figs. 10
and 11 for obtained positive as well as negative defects,
respectively. Here, WT and SP represent wiring track
segmented image and soldering pads segmented image,
respectively for template image. WT1 serve segmented
wiring track image of testing image (Fig. 8) (a).
Fig. 11. Classification of wiring track defects (negative)
B. Defects Related to Soldering Pads
Positive as well as negative defects analogous to soldering
pads are depicted in Table III. Under and Over etch
defects have larger area (~2000) compared to the area of
spur and mouse bite defects (~400). Adopting this
difference in area soldering pad defects are classified as
shown in Fig. 12.
C. Defects Related to Holes
Positive and negative defects related to holesare depicted in
Table IV. Bold fonts in Table IV performs shape of the defect.
TABLE III
DEFECTS RELATED TO SOLDERING PADS
Positive Defects (PDS) Under etch, Spur
Negative Defects (NDS) Over etch, Mouse Bite

Fig. 12: Classification of soldering pad defects
TABLE IV
DEFECTS RELATED TO HOLES
Positive Defects PDH Pinholes (Circle), Wrong
size (Big) hole (Ring) and
Breakout (Half-moon)
Negative Defects NDH Missing holes (Circle),
Wrong size (Small) hole
(Ring) and Breakout (Half-
moon)
(a) Circle shaped defect
(b) Ring shaped defect
(c) Half-moon shaped
defect
Fig. 13: Hole defects shapes
Fig. 14: Result generated by the proposed algorithm
There are mainly three shapes recognized in hole defects:
(1)circle(2)ring and (3)half-moonas shown in Fig.13.To
make the classification process invariant to rotation,
angular position and scale, Hu’s 2nd invariant moment
[17] is used to differentiate these shapes. Hu’s 2nd
moment for circle, ring and half-moon shapes are 3×10−5,
40 × 10−5 and 6390 × 10−5, resp.
TABLE V
TIMIMG REPORT OF THE PROPOSED ALGORITHM
Step Time (s)
Registration 1.143
Preprocessing 0.223
Defect Detection 0.001
Defect Classification 1.161
Total 2.528
VII. RESULTS
The final result gathered after classification step is depicted
in Fig. 14. It is observed that all the defects are successfully
disclosed and classified into correct classes. In addition to
this,the proposed algorithm takes just 2.528 s to executethe
investigation of a PCB image. The complete timing data for
each step of algorithm is explained in Table V. In the
proposed approach, except soldering pad defects, the
prospective algorithm uses scale invariant parameters (e.g.
number of connected component and shape-based moment
of defect) instead of using scale-based parameters like area
of defect. Scale invariantfeaturesmakeclassification process
robust to defect severity.
4
2
1
6
13
12
11 10
5 8
3 8 7
7
3
2 5 2
9 13 9
2
14 14
5 5 2

VIII. CONCLUSION
In this paper, I have proposed a novel method to disclose
and classify all available 14 types of defects of PCB using
referentialinvestigationmethod.Uniquenessofthealgorithm
is that it classifies all type of defects which is robust to defect
appearance and severity. The testing (defective) image is
coordinated with thetemplate (standard) image usingimage
registration techniques. Noise in the image is reduced with
help of process of median filtering and hence increasing
reliability and efficiency. Further- more, Gaussian low-pass
filtering is used in order to evadeuneven binarization due to
sharp transitions present at edges. The PCB image is divided
in three parts: wiring tracks, soldering pads and holes in
order to evaluate defects in different partsof PCB image. The
defect is disclosed using two-step process:imagesubtraction
followed by area filtering to eliminate small areas after
subtraction. After disclosingdefects, each defect is classified
using various region properties like number of connected
components, shape-based descriptors and area.
The prospective algorithm is able to identify all 14 types of
PCB defects, which is not covered in the state-of-the- art
algorithms. Also, prospective method takes only 2.528 s to
investigate a PCB image which makesitmoresuitableforAVI.
The algorithm is useful in electronics manufacturing
industries to investigatePCBquicklyandaccurately,thatmay
lead to reduced manufacture time and improvement in
overall efficiency, robustness and reliability of product.
REFERENCES
[1] R. T. Chin and C. A. Harlow, “Automated visual
inspection: A survey,” IEEE Transactions on Pattern
Analysis and Machine Intelligence, no. 6, pp. 557–573,
1982.
[2] R. T. Chin, “Automated visual inspection: 1981 to
1987,” Computer Vision, Graphics, and Image
Processing, vol. 41, no. 3, pp. 346–381, 1988.
[3] M. Moganti, F. Ercal, C. H. Dagli, and S. Tsunekawa,
“Automatic pcb inspection algorithms: a survey,”
Computer vision and image understanding, vol. 63, no.
2, pp. 287–313, 1996.
[4] W.-Y. Wu, M.-J. J. Wang, and C.-M. Liu, “Automated
inspection of printed circuit boards through machine
vision,” Computers in Industry, vol. 28, no. 2, pp. 103–
111, 1996.
[5] S. H. I. Putera, S. F. Dzafaruddin, and M. Mohamad,
“Matlab based defect detection and classification of
printed circuit board,” in Digital Information and
Communication Technology and its Applications
(DICTAP), 2012 Second International Conference on.
IEEE, 2012, pp. 115–119.
[6] T. Nakagawa, Y. Iwahori, and M. Bhuyan, “Defect
classification of electronic board using multiple
classifiers and grid search of svm parameters,” in
Computer and information science. Springer, 2013, pp.
115–127.
[7] S. Ren, L. Lu, L. Zhao, and H. Duan, “Circuit board
defect detection based on image processing,” in Image
and Signal Processing (CISP), 2015 8th International
Congress on. IEEE, 2015, pp. 899–903.
[8] S. Kumar, Y. Iwahori, and M. Bhuyan, “Pcb defect
classification using logical combination of segmented
copper and non-copper part,” in Proceedings of
International Conference on Computer Vision andImage
Processing. Springer, 2017, pp. 523–532.
[9] R. C. Gonzalez and R. E. Woods, Digital image
processing (3rd Edition). Upper Saddle River, NJ, USA:
Prentice-Hall, Inc., 2006.
[10] E. Rosten and T. Drummond, “Fusing points and
lines for high performance tracking,” in Tenth IEEE
International Conference on Computer Vision (ICCV’05)
Volume 1, vol. 2. IEEE, 2005, pp. 1508–1515.
[11] H. Bay, T.Tuytelaars, and L. VanGool, “Surf:Speeded
up robust features,” in European Conference on
Computer Vision. Springer, 2006, pp. 404–417.
[12] C. Harris and M. Stephens, “A combined corner and
edge detector”. In Alvey Vision Conference, vol. 15, no. 50.
Citeseer, 1988, pp. 10–5244.
[13] S. Leutenegger, M. Chli, and R. Y. Siegwart, “Brisk:
Binary robust invariant scalable keypoints,” in Computer
Vision (ICCV), 2011 IEEE International Conference on.
IEEE, 2011, pp. 2548–2555.
[14] D. Nistér and H. Stewénius, “Linear time maximally
stable extremal regions,” in European Conference on
Computer Vision. Springer, 2008, pp. 183–196.
[15] J. Shi et al., “Good features to track,” in Computer
Vision and Pattern Recognition, 1994. Proceedings
CVPR’94., 1994 IEEE Computer Society Conference on.
IEEE, 1994, pp. 593–600.
[16] P. H. Torr and A. Zisserman, “Mlesac: A new robust
estimator with application to estimating image
geometry,” Computer Vision and Image Understanding,
vol. 78, no. 1, pp. 138–156, 2000.
[17] M.-K. Hu, “Visual pattern recognition by moment
invariants,” IRE Transactions on Information Theory,
vol. 8, no. 2, pp. 179–187, 1962.

IRJET- Automatic Identification, Analysis and Investigation of Printed Circuit Boards for Defects and Errors Disclosure and Classification based on Nature

More Related Content

What's hot (20)

Similar to IRJET- Automatic Identification, Analysis and Investigation of Printed Circuit Boards for Defects and Errors Disclosure and Classification based on Nature (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Automatic Identification, Analysis and Investigation of Printed Circuit Boards for Defects and Errors Disclosure and Classification based on Nature