IJCSNS International Journal of Computer Science and Network Security, VOL.17 No.

6, June 2017 137

The implementation of automated optical inspection in printed

circuit boards
Arash Aravand
.Department of Electronic, Faculty of Electronic & Computer Engineering, Tabriz University, Parlar Co, Tabriz, Iran.

Jafar Sobhi
Department of Electronic, Faculty of Engineering, Urmia University, Parlar Co,Tabriz, Iran.

Abstract great progresses being made, current circuits are more

Different machines are used in the production of electronic complex than the boards being made a few years ago [2].
boards to control the elements, the correct amounts and their The introduction of SMT, and then the reduction of the
directions. The automated optical inspection systems are used to size of elements mean that the boards are compacted,
solve the excessive complexity of printed circuit boards and peace by peace. Even the relatively average-sized boards
increase the production with high accuracy as well as low errors.
In cases where the quick inspection of the board is impossible,
contain thousands of soldered nodes (points) and these are
the inspection is done visually with eyes. Since, the eyes get the points where most of the problems are caused [3] [4].
tired of continuous work, automating this process increases the In addition, these increasing complexities in boards will
production and quality of products. Therefore, a lot of visual mean that nowadays mechanical inspection is not
devices are presented in the industry. The automated optical appropriate. Today's market demands products with high
inspection systems can be used at any stage of production. AOI volume and quality being brought to the market in a fast
is one of these systems. In this technique, AOI machine is able to and reliable wat. Quick methods are needed to ensure that
get a picture of the board. The automated optical inspection product quality remains high. AOI,[2] automatic optical
system uses the registered image to compare the information of inspection, is an essential tool in in-circuit test ensuring
images with the information of the machines to detect errors and
that costs are kept low as much as possible through the
make decisions. Using this comparison, the AOI system can
detect and identify any error or suspicious area. In the present initial detection of errors. Automated optical inspection of
article, we have attempted to build an AOI device to cover all the printed circuit board is important because of the needs of
errors happening in the printed circuit boards at any stage of the modern manufacturing environment.[2][3] In the mass
production line. The traveling salesman problem is solved to production of electronic equipment, it has been tried to
control the direction of movement of the camera on the conveyor. increase the quality assurance factor in all the parts,
To introduce any printed circuit board to the system, a software components and finished products to 100%. This progress
is designed that uses CAD file to obtain the type of elements on has caused the electronics industry to use the automated
the printed circuit board and their conditions, finds the optimal visual inspection in manufacturing printed circuit boards;
route for the movement of X-Y table, detects the errors arising
board inspection may contain the identification of other
from lack of elements, elements' directions, the connection of
two bases, the lack of soldering, cold soldering, added soldering, false parts on the board, un installation of the pieces on the
etc. in three stages of feature extraction, feature selection and identified points, the false direction of pieces on the board,
decision-making. The results show that the device is efficient in checking the correct connection of the soldering, checking
the detection of the glue error before mounting of the element the interruptions of copper on the board as well as the
and the detection of SMD and DIP elements after bath. connection of two coppers to each other. In each case,
Keywords: suitable x-y tables, lighting equipment, and various
Evolutionary algorithms, artificial intelligence, feature selection, cameras are required. The quality of each of these
feature extraction, error detection components depends on the piece and the measurement
distance. With regard to the issues raised above, the
1. Introduction objective of this project is to detect the errors of the
printed circuit boards in production lines using image
Automatic optical inspection, is a key technology used in processing algorithms and artificial intelligence. Since,
the manufacturing of printed circuit boards. AOI provides most of these errors are not simply detectable by eyes and
a fast and accurate review of electronic collections and the visual inspection is very time consuming, error
guarantees them in certain PCBs; in this case, the quality probability is very high. That's why machine vision is used
of products is high and the elements are correctly and to increase the speed and reduce the errors.
flawlessly generated in the production line [1]. Despite the

Manuscript received June 5, 2017

Manuscript revised June 20, 2017
138 IJCSNS International Journal of Computer Science and Network Security, VOL.17 No.6, June 2017

In the field of error detection of SMD and DIP elements, The first section: Image Acquisition system
the following researches have been accomplished: a
method was provided by G. Acciani, G. Brunetti and G. a 5 Mpixel CDD camera with high resolution, a 12 mm
Fornarelli in 2006. In this detection, first, the IC image lens and an RGB lighting system has been used- being
and all bases of IC are removed from the board image, shown in figure 1. The working distance of the lighting
through Segmentation. Then, errors are detected by systems from the printed circuit board is 25 mm, the
removing Geometric and Wavelet features as well as distance between the camera and the board surface is 100
neural network training [6]. Fupei Wu and Zhang Xianmin mm. The amount of camera's FOV is 35 * 45 mm. To
proposed a method in 2011 that was used to detect IC cover the surface of the printed circuit board for shooting
errors through a lighting system with Red, Green and Blue an X_Y system was built with conveyor Figure 2.
lights. No Solder ،Surplos Solder ،Lacking Solder ،Lead
Lift ،Lead Left ،Shift ،Bridge ،Pesoudo Joint errors were
detected. The type of error is determined by feature
extraction and segmentation of images and comparing
them with a set threshold values [5]. A method is provided
concerning the error detection in resistors and capacitors
by Xie Hongwei et al. in 2011. The RGB lighting system
was used for the detection; this method, as the previous Figure 1.RGB lighting system
method has used element segmentation. A set of features
is extracted for each segment; due to the high number of
features, the most effective feature as well as a method is
proposed for each segment. Using the decision tree, the
appropriateness or inappropriateness of the segments are
determined. [7] To detect the basic soldering error of DIP
elements, T.S. Yun and colleagues proposed a method in
1999. The method has used the removed features' image to
detect errors and has provided a backup vector for each
type of error through the use of X feature vector. [9] B. C.
Jiang et al. in 2006 proposed a method for error detection
with the removal of printed circuit boards' background and
Figure 2. X-y table and conveyor
the discovery of basic parts through the feature extraction
and comparing it with threshold values. Good Solder,
Open Solder, Bridge, No Solder errors were detected. [8]
The aim of this project is to build a device to detect errors
The second section: Inspection Algorithm
before and after installation of elements on the board. The This section detects errors associated with 1) SMD
device has the ability of error detection glued on the resistance and capacitors, 2) IC SMDs, 3) basic elements
printed circuit board, before the element installation. Also, in the DIP, 4) the detection of glue error on the board, 5)
after mounting elements, the device is capable of error The best path to solve the traveling salesman problem. The
detection for SMD elements and DIP basic elements on following operations are conducted in order to detect
the surface of printed circuit board. Therefore, a software errors in all sections of 1 to 4; image segmentation, feature
is designed to receive the position of elements or segments extraction, feature selection and decision-making. These
of the printed circuit board that need to be inspected by the stages are explained in every segments.
user or through CAD file. Then, the program being based The image of each element is divided into a series of
on the position of defined elements extracts the centers of segments after the removal of the original image of the
shooting points and solving the traveling salesman board. These segments differ based on the type of
problem, it selects the best shooting path for the device. elements and errors. In this study, the segments are
Finally, the device is ready for shooting and error combined according to the type of error being detected for
detection. In the first part, the hardware of the device has the elements. In fact, by combining values of existing
been described. In the second part, the procedures of error properties, we have created fewer features in a way that
detections of elements are mentioned along with their these features have all the basic information in the initial
results. Finally, conclusions and comparisons with other features. In this part of the segment which is an RGB
works are expressed. image, first the desired color is removed from the desired
ranges (red, blue, green and the color of elements) using
the Color Segmentation. A series of removed colors are
IJCSNS International Journal of Computer Science and Network Security, VOL.17 No.6, June 2017 139

used to obtain some information about the error detection, B) Feature Extraction:
this information is called features. Since each image in Features extracted from R1 to R5 segments for the RGB
RGB color space has 3 gray images image in each segment are:
and for each image 3 binary images can be calculated, the 1. The red, blue, green area in every segment
number of features for each segment -taking into account 2. The conjunction of all three colors in the length
the extraction of 10 features- is 30 features (10*3 for each and width of each segment
segment). Thus, the features in this sections are discussed. 3. Calculation of the center of mass of every 3
The issue of feature selection, is one of the issues being colors in each segment
discussed in the context of machine learning and statistical 4. Calculation of the maximum amount of
identification of patterns. This issue is of great importance continuity
in many applications, because there is a large number of 5. The difference between the largest and smallest
features in these applications, many of whom are either gray value for all 3 dimensions in each segment
unused or of little information.[10] It does not create a 6. The average of gray features' design
problem in terms of information if the features are not 7. Average gray images for each 3 dimension
removed, but the computational load for a given 8. Objects' being in the shape of squares and circles
application increases.[10] In addition, it makes a lot of for each 3 dimensions in each binary segment
useless information to be stored along with useful data.
Due to the increase in features, we have attempted to find
the less features with the most effects in the resistance and C) Feature selection for R1 to R5 segments:
capacitance part using genetic multi-objective algorithm The principles for solving this problem follow the same
NSGA-2. general principles stated in the multi-objective genetic
algorithms [13] [14]. Here, we just refer to the methods
A) Resistance and SMD capacitors used in the project and explain the steps involved in the
The elements of this group have two bases on the sides project as well as the way of performing these steps.
and the element is placed in the middle of the pad. Figure Encoding of feature selection is a binary encoding that
3 shows the way of segmentation of SMD capacitor's pads, aims at converting the answer to a series of binary
body of the element and sides of the element. To detect numbers (in binary). Crossover in the feature selection: 1.
errors in the group first as shown in Figure 3. Each base is Single Point Crossover method, 2. Double Point Crossover
divided into a series of segments. The features mentioned method and 3. Uniform Crossover method are selected.
in the section on feature extraction are extracted in the Production phase of the initial population: according to the
section feature extraction, are extracted for each of the type of encoding system, the chromosomes of the initial
segments of R1 to R5. Due to the large number of features population are produced for the searching. The number of
and segments for detection, the best (most effective) the population, the chromosomes or the population size
features are selected using genetic multi-objective depend on the number of demanded population by the user.
algorithm NSGA-2. In the following, we will explain and The initialization phase: in this phase an initialized
discuss the results. After the feature selection, the function is used calculate the cost of each chromosome;
decisions are made about features by the cost faction of this phase is the most important phase in the genetic
the evolutionary algorithm. Finally, to detect the type of algorithms. Two factors are used in this project. One of
errors, the outputs of each segment are combined and by these factors is caused by the number of features and the
the use of fuzzy logic the type of error has been other factor is the cost function of the chromosome
determined. This is done intuitively by setting a Mamdani categorization. Since, the aim is to reduce the error rate as
fuzzy logic. Further explanation of R6 and R7 segments well as the features, this issue is solved as a multi NSGA-2
are given below. issue. As described in the multi-objective genetic
algorithms above [14][15] one advantage of this algorithm
is that it gives us a series of answers for decision-making
as Pareto Front at the end of each phase. Evaluation phase:
In this phase, the evaluation is on the classification of
Pareto Fronts according to the priorities which are selected
according to the number of the populations. Taking a ratio
for the number of features in each stage, we can search for
the best answer among the Pareto Fronts with priorities of
one. The stopping criterion for the algorithm is that if an
Figure 3. the elected segments of the SMD capacitors' pads for error
answer is repeated, first the Mutation factor in algorithm is
detection
increased so that in case of being trapped in the local
140 IJCSNS International Journal of Computer Science and Network Security, VOL.17 No.6, June 2017

minimum, it can get out of it, but if an answer is repeated are mentioned in the following. Table 3. Shows the error
more than a certain amount, the algorithm is stopped. The detection results for 1000 capacitor and resistor elements
flowchart is shown in Figure 4. The cost function: two used in 4 different boards. The same applies to error
cost functions are used for feature selection; LVQ faction detection methods of other SMD elements such as diodes
and the combination function being combined of some and LEDs of SMDs.
SVMs. Each categorizing function, categorizes each Coefficients used in the algorithm:
segment of the pad into three categories: 1. Good 2.
Excessive 3. Bad. The procedure was performed for each 2. Roles:
segment of 400 training data; 40 percent of them were
used for training of the categorization faction and 60 1- If (R1) is (Excessive) and (R2) is (Excessive) and (R3)
percent were used for the Test function. The best cost, is (Excessive) and (R5) is (Excessive) Then Pad is
error and effective features of the results of both cost (Excessive solder)
functions are shown in Table 1. In the following, a fuzzy 2- If (R1) is (Good) and (R2) is (Excessive) and (R3) is
mamdani logic is intuitively set for the decision making to (Excessive) and (R5) is (Excessive) Then Pad is
decide on the error by giving cost function's qualities in (Excessive solder)
every 5 segment (good - bad - Excessive). Figure 5 shows 3- If (R1) is (Bad) and (R2) is (Excessive) and (R3) is
a block diagram of fuzzy logic to determine the type of (Excessive) and (R5) is (Excessive) and (R4) is
error. Finally, 5 types of error out of 5 segments are used (Excessive) Then Pad is (Tumbstone)
in the soldering of Resistance elements and SMD 4- If (R1) is (Bad) and (R2) is (Bad) and (R3) is (Good)
capacitors. This 5 types of error can be named as follows: and (R5) is (Bad) Then Pad is (Cold solder)
No soldering, Excessive soldering, No pad (tomb stone), 5- If (R1) is (Bad) and (R2) is (Bad) and (R3) is (Bad)
cold soldering and safe soldering. For more explanation on and (R5) is (Bad) Then Pad is ( NO solder)
the way of detecting errors by the fuzzy logic, a few rolls

Table 1 Results of feature selection for individual segments

Regions R1 R2 R3 R4 R5
2.0907 4.7801 3.1127 0.5702 0.4883
LVQ
7 5 5 3 5
2.5281 4.3944 3.3708 0.987 0.8424
SVM
7 6 5 4 9

Table 2 coefficients used in feature selection algorithm

The number of repetitions The initial population Pressure coefficient Recombination rate Mutation rate
100 100 0.8 0.5 0.7

Figure 4. Flowchart of the feature selection

IJCSNS International Journal of Computer Science and Network Security, VOL.17 No.6, June 2017 141

The sixth segment R6:

The sixth segment is taken into account for the detection
of 'no element' error. Removing the color of the element's
body and comparing it with a threshold value, we can
decide on the detection of the 'No element' error. The
threshold varies for the removal of the desired color by the
user.

Figure 5. The way of dividing segments of IC

1) SMD ICs:
This group, like the first group is divided into a series of
segments. For each segments, some features are extracted
and finally the error is detected for each segment. The
Figure 5. a block diagram of Mamdani fuzzy logic segmentation of ICs for error detection is shown in Figure
5. The bridge, Pseudo joint, no element and element
direction are detected for ICs.
3. The seventh segment R7: For the error detection of ICs, first R4 segment is
investigated, if there is no error in this segment, other
errors will be investigated. If an error is detected in the R4
The seventh segment is considered for the detection of
segment, it is considered as No element error and there no
bridge error between two pads. Removing the total area of
need to investigate the other segments. 4 segments are
the solders' color (Red, Green, Blue) and comparing it
taken into account for ICs and since these segments are
with a threshold value, we decided on the detection of the
adequate for the type of error detections, and each segment
bridge error between two pads.
is determinant of one type of error, also the segments are
Table 3. Results of the error detection of SMD resistance and capacitors not dependent on each other like the first group, less
The type The number and features were involved in the error detection. Therefore,
of The type of error percent of the feature selection stage has been removed from this
categorizer respondents error detection. In the following, we will describe the segments
No solder 97 91 93.81 and selected features for each segment. The first R4
LVQ & Excessive solder 377 365 96.81 segment indicates pseudo joint error. First, the area
mamdani Tombstone ( No pad ) 5 5 100 occupied by Red, Green, Blue colors is extracted. If the
Fis Cold solder 21 19 90.47 red color is more than the sum of the two other colors, the
Safe solder 478 473 98.95 segment has a pseudo joint error. The second segment R2
No solder 97 95 97.93 indicates the Bridge error. First the area occupied by Red,
SVM & Excessive solder 377 365 96.81 Green, Blue colors is extracted. Calculating the largest
mamdani Tombstone ( No pad ) 5 4 80 association of colors for Red ،Green ،Blue and comparing
Fis Cold solder 21 18 85.71 both features of the segment, it is realized that the error
Safe solder 478 452 94.56 can be detected if the features are larger than the threshold
R6 No element 18 17 94.44 value of R2. Usually ICs have a series of symbols to detect
R7 Bridge between two gaps 4 4 100 their direction such as small circle on ICs or semicircle at
the top of the body of IC or a line on the body. These signs
are used in the R3 third segment to detect the direction of
IC. Segment 3 specifies the right direction of ICs;
extracting the colors from this segment and comparing it
with the threshold value, we can detect the right direction
of ICs. If the resulting number is less than the threshold
value, the error is detected for the segment. The R4
142 IJCSNS International Journal of Computer Science and Network Security, VOL.17 No.6, June 2017

segment identifies the No element error. By removing the The features related to the area of the Red color as well as
color of IC's body – which is black- or the green color of the dark color of the holes in bases are extracted.
the printed board and comparing it with the threshold Calculating the area and comparing it with the threshold
value, the No element error is detected. Table 4 shows the value, the error is detected. If the condition is not met, the
results of the 800 different ICs in 4 different types of Hole error will be determined. The threshold value is
boards: changeable by the user. As seen in Figure 7, several
features are combined to make decisions. Table 4 shows
Table (4) the results of error detection for ICs of SMD the results of the detection for 990 DIP pads from 4 boards.
The number of errors and the Dis: shows the distance of the desired color from the
Type of errors
percentage of error detection center of the image.
Bridge 200 199 99.5 Cir: shows the circularity of the desired color
Pseudo joint 200 194 97 Area: shows the area of the desired color in the image
No element 40 40 100 BigArea: shows the largest area of the object in terms of
Direction of
40 38 95 the desired color.
the element
Total percentage 98.875

2) DIP elements:
Some features are extracted from the image of the pad for
the third group or DIP elements. Considering 3 main
segments for the pads – being shown in figure 6- the errors
of ،No Solder ،Less Solder ،Bridge، ،Holes in Solder,
Excessive Solder, and no lead, safe solder are detected.

Figure 7. The flowchart of error detection for the second segment of Dip

Table 4. Results for the basic elements of DIP error detection

Type of errors Number of Number of correct Output
errors detections
Bridge 212 212 100
Excessive 50 55 100
solder
Less solder 85 81 95.29
Figure 6. Segmentation of DIP elements Hole in the
solder 178 162 91.01
The first segment which detects the Bridge error is like R2 No solder 8 7 87.5
No solder 43 47 100
segment in ICs. The three other errors occur in the second Safe solder 414 411 99.27
segment R2and as in resistance, the capacitors need high Total percentage 96.152
precision for the detection. This precision is because of the
presence of large number of errors in a segment. As with 3) Glue detect errors in printed circuit boards:
all segments, we extract the features from the segment and
detect errors of less solder, excessive solder, No pad, and SMD elements are assembled on printed circuit boards
No solder. The following feature are extracted; the area of either through Paste or tin bath. In the second method, Fig.
Red, Green, Blue colors in the segment, the largest object 8 and 9 Red Glue is used for the fixation of the elements
for the three colors of Red, Green, Blue of the segment, on the printed circuit board and avoidance of falling into
the largest association of extracted color in length and tin bath. In order to avoid errors such as dropping of the
width of the segment, the circularity amount of the closest elements, the elements' shift on the board and the
red color to the center of the base and the center of the dislocation of glue under the element, we have discussed
mass of object in segment. Threshold value for the error Glue errors before Mount operation. In this section we
detection in each segment or blocks can be changed by the have tried to detect errors of glue's wrong location,
user. The third R3 region is located right in the middle of rubbing of the glue, the inadequateness of the volume of
the second segment containing the 50% of the second the glue, the lack of glue and safe glue.
segment; it is used to detect the error of Holes in the solder.
IJCSNS International Journal of Computer Science and Network Security, VOL.17 No.6, June 2017 143

1- If GlueAreaLess and GluePeremeterLess Then

LessGlue
2- If GlueCircularityLess and GlueCountinuslyWidthLess
and My_Bad Then RubbingGlue
3- If GlueCircularityLess and
GlueCountinuslyLengthLess and Mx_Bad Then
RubbingGlue
4- If Number_of_GlueObject_MoreThanOne and
GlueCountinuslyWidthMach Then RubbingGlue
5- If Number_of_GlueObject_MoreThanOne and
GlueCountinuslyLengthMach Then RubbingGlue

Figure (8) Image of the glue with two lighting systems

The improvement of the moving path of the camera
to reduce the time processing of shooting:
The whole time of the process is started from the shooting
time on the printed circuit board and continues till the
detection of all errors of the board. To reduce the
travelling time of the device and the suitable route for
shooting, we need to select an optimal rout with the
shortest length.[11][12] This problem can be solved by
connecting the rout of the camera to the Traveling
Salesman Problem and solving it.
Solving the Traveling Salesman Problem with genetic
algorithm
In this section we will discuss the way of solving the
Traveling Salesman Problem in this project. The encoding
system used in this project is of permutation kind of
Figure 9. the image of the board before Mounting
encoding known as an encoding. According to the type of
encoding system, the initial population of chromosomes
After shooting and extracting the check points as previous are produced for searching. The population number or the
methods, some features are removed from the image for number of chromosomes to search or the Population Size
error detection; also fuzzy logic is used to classify the type depend on the requests of the user. The initialization phase
of errors. Extracted properties are: 1. Area of the glue in in which an initialization function is used to calculate the
the image (the color of the glue), 2. Perimeter of the glue fitness of each chromosome is one of the most important
in the image, 3. The circularity of the glue, 4. The Max phases in any genetic algorithm. Two factors are used in
and Min extraction of the object continuity in the image as the project. One of these factors is caused by the number
glue, 5. Extracted object number as glue, 6. Calculation of of edges in the permutation of chromosomes and the other
the center of extracted mass area as glue, 7. The area of factor is the length of the rout for the edges of
the blue part of the glue in the picture is shown with the chromosome. Since we want to join less features with
blue color. In this survey, the volume of the solder in not more weights for each chromosome, we realize that the
expressed numerically. The error is only detected if the chromosome joining less features with more weights has a
glue does not have the right amount to keep the element; greater fitness and the likelihood of their selection in
the error is shown to the user on the image. Mamdani genetic operators becomes more. The population
fuzzy logic is used to detect. For example, some samples production phase includes Mutation and Crossover. In this
of rolls are given below regarding the fuzzy logic: project, like the feature selection method, we can use
random selection or RouletteWheel selection. The
Table (9) Results of the glue error for 3500 different boards RouletteWheel selection system, first locates all the fitness
No glue 100 chromosomes on the RouletteWheel and then one of them
Inadequate volume of the glue 91 is selected through random selection. The advantage of
The rubbing of the glue 97.61 this method is that the chromosome having more fitness
Inappropriate location of the glue 98.7 occupies more sectors than RouletteWheel and is likely to
be selected. The Crossovers such as PXM and OX are
Less glue 93
used. A possibility is considered for each of them being
The total percentage of the glue 96.062
144 IJCSNS International Journal of Computer Science and Network Security, VOL.17 No.6, June 2017

randomly selected by RouletteWheel, Crossover. The

crossover which has a greater probability is more likely to
be selected. Mutatios such as: Insert Mutation, Swap
Mutation, Inversion Mutation, are used in this process. A
possibility is considered for each of them being randomly
selected by RouletteWheel, Mutation. Mutation that has a
greater possibility algorithm has a better chance of
selection. In this project, we have considered the
evaluation phase on the ground that at any moment an
optimal tour is found, the optimized weight of the tour is
given to the user so that the user can stop the search
algorithm at any time. The stopping condition for the Figure (11) the optimal route for the second sample
algorithm is that if an answer is repeated, first the
Mutation factor in algorithm is increased so that in case of The improvement of the fitness function over iteration:
being trapped in the local minimum, it can get out of it,
but if an answer is repeated more than a certain amount,
the algorithm is stopped. The process is implemented for
further comparison in two different directions by changing
the initial population and the number of repetitions. The
results are shown Figure 10 shows a flowchart of Genetic
Algorithm.
According to the earlier sections we can say the power of
genetic algorithms on the processors is parallel. This
power on one processor does not show a good
performance. In addition, the number of searching genetic
algorithms affect the speed to achieve the optimal answer.
The higher the number of the population is, the greater is Figure (12) the improvement of the fitness function over iteration for the
the speed of reaching an optimal solution. Since the second sample
genetic algorithms are suitable for the searches with large
ranges, the large number of chromosomes in question lead B) sample:
to the gradual reduction of the convergence rate. So with
Table (8) Results of the TSP problem for the third sample
regard to the raised issues it is suggested to use the
molecular coding to improve the movement of the camera; It = 500 It = 300 It = 100 Population
because it implements genetic structure. NFE =75100 NFE =45100 NFE =15100 100
55.6810 57.3902 84.08
Table (6) coefficient of genetic algorithm for solving TSP NFE =66900 NFE =135300 NFE =45300 300
Pressure 55.3250 55.9601 68.1815
Mutation rate Recombination rate
coefficient NFE =135500 NFE =92750 NFE =75500
0.7 0.8 10 500
56.3284 56.7408 77.5184
A) Sample: The optimal route:

Table (7) Results for solving TSP for the second sample
It = 500 It = 300 It = 100 population
NFE =
NFE =45100 NFE =15100
75100 100
168.5805 263.7227
156.1930
NFE =
NFE 135300 = NFE =45300
66900 300
165.669 266.737
156,7200
NFE =
NFE9.2750 = NFE =75500
135.500 500
162.287 282.935
154.4557
Optimal rout: Figure (13) the optimal route for the third sample

The improvement of the fitness function over iteration:

IJCSNS International Journal of Computer Science and Network Security, VOL.17 No.6, June 2017 145

Defect [7] [5]

GS Y Y Y
NS Y Y Y
SS N Y Y
LS N Y Y
PJ Y Y Y
CS Y N N
ES Y N N
TS Y Y N
MS Y N N
Br Y N Y
SF N Y N
Dir Y N N
LL N N Y
Sh N N Y
Figure (14) the improvement of the fitness function over iteration for the LB N N Y
third sample SVM & Mamdani & IC LVQ & Mamdani & IC 98.6
Total 97.3
95.32 97.12

4. Conclusion:
Our goal in this project was to construct AOI system Fig.
15 and 16, and create algorithms for detecting various
errors on various elements, being used in all sectors of
production line. As mentioned the built device and its
software can be used in two production lines, before
mounting the elements and after the solder bath. In
addition, with minor changes the mentioned methods can
be used to detect errors of other elements such as:
regulators, diodes, LED, SMD, and transistor. Tables (10
and 11) indicate a comparison between our method and
earlier works. It is realized from the above tables that GS ،
NS ،SS ،LS ،PJ ،CS ،ES ،TS ،MS ،Br ،SF ،Dir ،LL ،Sh ، Figure (15) Inside of the built machine

LB ،NP ،HS and ES have the errors of Good solder ،No

Solder ،Surplos Solder ،Less Solder ،Pseudo Joint ،Cold
Solder ،Excessive Solder ،Tomb Stone ،Missing ،Bridge ،
Shift ،Direction ،Lead Left ،Shift ،Lead Bend ،No Pad ،
Holes in Solder and Excessive Solder, respectively. As it
can be seen, the number of errors detected in this work are
more than the previous works. The comparison is
indicated in the following tables. The identification of the
glue error is not considered in the tables, because it had
not been regarded in the previous works.

Table 10. Comparison of the basic elements of DIP

Type Defect Our Method SVM [9] CHF [8] Figure (16). The built AOI machine
GS Y Y Y
ES Y Y N
LS Y Y N
HS Y N Y
References:
[1] H. H. Loh and M. S. Lu, “Printed circuit board inspection
NP Y N N
using image analysis,” IEEE Trans. Ind. Appl., vol. 35, no.
NS Y Y Y
2, pp. 426–432, Mar.–Apr. 1999.
Br Y N Y [2] M. Moganti, F. Ercal, C. H. Dagli, and S. Tsunekawa,
Total 96.152 98.01 97.3 “Automatic PCB inspection algorithms: A survey,” Comput.
Vision. Image Understand., vol. 63, no. 2, pp. 287–313, Mar.
Table 11 comparison of the resistance and IC methods 1996.
Type Our Method Decision SDL
146 IJCSNS International Journal of Computer Science and Network Security, VOL.17 No.6, June 2017

[3] J. D. Sim, F. S. Galbraith, and N. Davenport, “Modelling of

PCB variations and its impact on microcontroller
immunity,” in Proc. 10th Int. Conf. Electromagn. Compat.,
Coventry, U.K., Sep. 1997, pp. 100–104.
[4] P. D. Jadhav and J. S. Smith, “Analyzing printed circuit
board assembly lines using a PCB assembly template,” in
Proc. 37th Winter Simul. Conf., 2005, pp. 1335–1342.
[5] Fupei Wu and Xianmin Zhang . 5 May 2011. Feature-
Extraction-Based Inspection Algorithm for IC solder joints.
IEEE TRANSACTION ON COMPONENTS,
PACKAGING AND MANUFACTURING
TECHNOLOGY
[6] G. Acciani, G. Brunetti and G. Fornarelli. 2006. A Multiple
Neural Network System to Classify Solder Joints on
Integrated Circuits. International Journal of Computational
Intelligence Research.
[7] Zhang Xianmin and ets . 12 December 2011. Solder joint
Inspection Method for Chip Component Using Improved
AdaBoost and Decision Tree . IEEE TRANSACTION ON
COMPONENTS, PACKAGING AND
MANUFACTURING TECHNOLOGY
[8] B. C. Jiang, C. C. Wang, Y. N. Hsu. 3 July 2006. Machine
Vision and background remover-based approach for PCB
solder joints inspection. International journal of production
reserch. Taylor & Francis
[9] T.S Yung, K.J. Sim and H.J. Kim. 25th May 2000. Support
Vector Machine-Based inspection of solder joints using
circular illumination .[10]. M. Dash, H. Liu, 1997 . Feature
Selection for Classification. Intelligent Data Analysis
1:131-156,.
[10] Peter G. Anderson. 2004. Permutation Based GAs and
Ordered Greed. Computer Science Department Rochester
Institute of Technology, Rochester, New York,
[11] John LaRusic. 2004. An Introduction to the Traveling
Salesman Problem.
[12] Peter G. Anderson. 2002. Introduction to Genetic
Algorithms. Computer Science Department Rochester
Institute of Technology, Rochester, New York,
[13] Kalyanmoy Deb, Samir Agrawal, Amrit Pratap, and T
Meyarivan. A Fast Elitist Non-Dominated Sorting Genetic
Algorithm for Multi-Objective Optimization: NSGA-II.
Kanpur Genetic Algorithms Laboratory (KanGAL) Indian
Institute of Technology Kanpur Kanpur, PIN 208 016, India
[14] Carlos A. Coello Coello Gray B.Lamont David A. Van
Veldhuizen, Evolutionary Algorithm for Solving Multie
Objective Problems . IlliGAL, Dept. of General Engineering
University of Illinois at Urbana-Champaign Urbana, IL
61801 USA