Sunteți pe pagina 1din 13

Faculty of Engineering

Dual Studies

Practical Period Report #VI

By: Mohannad Sehwail

ID#: 21211412

Supervisor: Dr.Mohammed Kouali

Jerusalem – Palestine
Academic year: 2019-2020

Table of Contents

`About Al – Quds University.................................................................................................................3

About Electro Co. LTD.........................................................................................................................3

About the Dual Studies Program...........................................................................................................3

Overview of this Report:.......................................................................................................................4

Activities Achieved...............................................................................................................................8

Obstacles:..............................................................................................................................................9

Lessons Learned:.................................................................................................................................10

Recommendations:..............................................................................................................................10

Conclusion...........................................................................................................................................10

References…………………………………………………………………………….…………………………………………………………12

2|Page
About Al – Quds University
Al-Quds University (The Arab University in Jerusalem) is a non-profit public higher
education institution located in the rural setting of the large city of Jerusalem called Abu Dis
and is officially accredited by the Ministry of Education and Higher Education. Al-Quds
University is a medium-sized coeducational institution which offers courses and programs
leading to officially recognized higher education degrees such as pre-bachelor degrees (i.e.
certificates, diplomas, and associate or foundation degrees), bachelor degrees, and master
degrees in several areas of study. The University also provides several academic and non-
academic facilities and services to students including a library, housing, sport facilities and/or
activities, financial aids and scholarships, study abroad and exchange programs, as well as
administrative services.

About (the IPV – University of Stuttgart)


Institute of Photovoltaics (IPV) was founded by professor Werner Kluge in the 1950s, and
until October 2011 known as “Institute of Physical Electronics”, the ipv is one of 24 institutes
of the Faculty of Computer Science, Electrical Engineering and Information Technology at
the University of Stuttgart.

The Institute for Photovoltaics researches and teaches with regard to the manufacturing,
characterization and use of materials, components, and systems in semiconductor electronics
and electrical energy storage systems; with a special focus on their use in renewable energy.
Further activities include defect-free laser processing of semiconductors and sensor
technology.

About the Dual Studies Program


The project "More Job Opportunities for Palestinian Youth" through Dual Studies at Al-Quds
University is the first model in Palestine to combine between the theoretical and practical
studies, assisted by the German Government, is being carried out by GFA Consulting Group
on behalf of the GIZ.

This program has been designed to give the Palestinian students the opportunity to raise their
professional level and gain knowledge to be effective in the Palestinian labor market.

3|Page
The Dual Studies program was launched in 2015, and offered a bachelor degree in Electrical
Engineering and Information Technology, in 2016, the program also offered a degree in
Business Administration.

Overview of this Report:


As this practical period was different from the previous ones, since this practical period was
an internship in the institute of photovoltaics, at the sensor technology department, this
practical period was rich of the new experience and activities. In this report we will discuss
the main activities achieved during this internship, the new technologies and methods of
working in electrical engineering in such a well-developed country as Germany and how to
apply this knowledge in the Palestinian Labor Market.

Introduction:
This internship in the Sensor technology department of the institute of photovoltaics was
about developing a software and hardware system to detect the defects in the solar modules
automatically, and the internship was divided into sub main tasks to reach our goal.
Task #I: Crack labelling using image labeler tool from MATLAB:
Image segmentation is a commonly used technique in digital image processing and analysis
to partition an image into multiple parts or regions, often based on the characteristics of the
pixels in the image. Image segmentation could involve separating foreground from
background, or clustering regions of pixels based on similarities in color or shape.
Task #II: Labeled Images analysis
The aim of this task, is to compare the accuracy and error of the human labeled images
using MATLAB image labeler tool, the same images were labeled by different employees.
The identified cracks in the solar cells, will be the outputs of the Neural Network (the final
task) which will be used to identify the defects in the solar cells automatically in the future,
and so, we wanted to study the effect of the human error on the labeled images, which could
also be used to further reduce the error in the next time.
Method: Image processing
The size of the labeled crack differs from on employee to another, based on the size of the
brush used to label the cracks, and so the first thing to be done, as a solution to this problem,
and to unify the size of the labeled cracks from different employees was to use something
called Skeletonization, in which the width of the crack is reduced to one pixel only.
The next step is to analysis those cracks, and to do this, I classified the cracks into 3 types:

4|Page
1- Separated crack (end point to end point crack)
2- Brach to Brach crack
3- Branch point to End Point Crack
Based on this classification, we developed python script to compare the cracks in each
labeled image for certain employees (e.g. Mohannad vs Pascal).
The comparison included (per each label):
1- Type of cracks.
2- Count of Cracks
3- Length of cracks in Pixels
4- Location of Cracks (starting and ending point of the cracks)
The final data was written an Excel sheet contain the previous data, which was used
later to discuss the differences in labeling between the employees.

figure(1): Comparison for the number of cracks per each label

Task #III: BUS BAR detection:


In our case, each solar cell contains 2 busbars, and the busbars separates part of the cell from
another, and so, the area between a crack and bus bar in the solar cell, is considered as a
closed area, so identifying busbars could help in identifying the defected areas in the solar
cells.
Method:

5|Page
At the pixel level, the bus bars are just two straight lines that extends from edge to the edge of
the solar cell image, and to save time and effort (we didn’t want to label them using the
MATLAB image labeler tool), we developed a software that could search for the busbars and
return them.

Hough Transform method:


Hough Transform is a feature extraction technique used in image analysis, computer
vision, and digital image processing. The simplest case of Hough transform is detecting
straight lines. In general, the straight line y = mx + b can be represented as a point (b, m)
in the parameter space. However, vertical lines pose a problem. They would give rise to
unbounded values of the slope parameter m. Thus, for computational reasons, Duda and
Hart proposed the use of the Hesse normal form

Task #IV: Defect labeling


Defect labelling of the solar cells is the final outcome of our project, but since we decided to
train the neural network to find the cracks and busbars only, we need a software to label the
defected areas in the solar cells based on the cracks and the busbars. The idea is very simple,
first we make sure that the busbars in the mask image (the labeled image using the MATLAB
image labeler tool), extend form the edge to the edge of the image, then we closed the gaps
for the cracks which end near the edge of the image within 5-pixel error.
After this, the closed areas between some 2 or more cracks are considered as a closed area,
and the area closed between the busbar and a crack or more also as a closed area, and
between a crack and the edge of the image as a closed area. The function then should look for
those closed area in the 2D array mask image, and return the coordinates of all the closed
areas in the mask image.
The next step is to look whether the closed areas are defected or not, and since we have the
coordinates if the closed areas, we can use the same coordinates to look for the mean
intensity in the original image and decide whether this area is defected or not.
We classified closed defected areas into three types: RGB (red, blue, green), red being the
most defected and green the least defected based on the intensity of that area.
Note: this software was developed to label the images labeled by MATLAB, but to use this
software to label the mask images generated by the neural network, we need to do some

6|Page
modifications of the software, for example, the neural network should return the busbars and
the cracks, so we don’t have to search for the busbars again in the image.

Task #V Neural Network Development:


Introduction to Neural Networks:
The term Deep Learning or Deep Neural Network refers to Artificial
Neural Networks (ANN) with multi layers. Over the last few decades, it
has been considered to be one of the most powerful tools, and has
become very popular in the literature as it is able to handle a huge
amount of data. The interest in having deeper hidden layers has recently
begun to surpass classical methods performance in different fields;
especially in pattern recognition. One of the most popular deep neural
networks is the Convolutional Neural Network (CNN). It takes this name from mathematical
linear operation between matrixes called convolution. CNN have multiple layers; including
convolutional layer, non-linearity layer, pooling layer and fully-connected layer. The
convolutional and fully- connected layers have parameters but pooling and non-linearity
layers don't have parameters. The CNN has an excellent performance in machine learning
problems. Specially the applications that deal with image data, such as largest image
classification data set (Image Net), computer vision, and in natural language processing
(NLP) and the results achieved were very amazing. In this paper we will explain and define
all the elements and important issues related to CNN, and how these elements work. In
addition, we will also state the parameters that effect CNN efficiency. This paper assumes
that the readers have adequate knowledge about both machine learning and artificial neural
network.

Method:
We developed 3 models of CNN, 54 layer NN, 67 and 103 layers NN, then we used the
collected data to train the neural network.
Input Images are called features
Labeled Images are called  labels

7|Page
The neural network will learn from the 3 labeled classes (crack, busbar and background) to
segment the images.
The parameters to be adjusted to train the neural network to get the best accuracy and reduce
the loss, are the learning rate, number of epochs and number of classes.
We trained the first model (67 layer) on 600 images/labels divided into:
Training set  60%
Validation set  20%
Testing set  20%
Accuracy and Loss measurement method:
Intersection over Union
The Intersection over Union (IoU) metric, also referred to as the Jaccard index, is essentially
a method to quantify the percent overlap between the target mask and our prediction output.
This metric is closely related to the Dice coefficient which is often used as a loss
function during training.
Quite simply, the IoU metric measures the number of pixels common between the target and
prediction masks divided by the total number of pixels present across both masks.
IoU=target∩predictiontarget∪predictionIoU=target∩predictiontarget∪prediction
As a visual example, let's suppose we're tasked with calculating the IoU score of the
following prediction, given the ground truth labeled mask.

Results:

8|Page
figure(2): Model Acc

The model shows that the accuracy is increasing with the number of epoch, epoch one
iterations of all the data set

9|Page
figure(3): Model Loss

figure(4): Example of segmented image

10 | P a g e
figure(5): Example of labeled defected cell, red most defected, and blue least defected

Obstacles:
The training period was very informative but the time was too short to gather all the
information in such a great topic.

Lessons Learned:
- How to use Machine learning to solve problems in Electrical Engineering.
- New techniques and methods of working in Electrical Engineering.

Conclusion
This practical phase was a shift in applying electrical engineering new sciences, using
machine learning and neural networks which is a very powerful tool, new instruments and
methods of working especially in the field of solar energy development were introduced in
this phase, and this exchange program could really change the way how the students look
towards the Electrical Engineering.

11 | P a g e
References:
1- Krizhevsky, A., Sutskever, I. and Hinton, G.E., 2012. Imagenet classification with
deep convolutional neural networks. In Advances in neural information processing
systems (pp. 1097- 1105).
2- Wei Xiong , Bo Du, Lefei Zhang, Ruimin Hu, Dacheng Tao "Regularizing Deep
Convolutional Neural Networks with a Structured Decorrelation Constraint ” IEEE
16th International Conference on Data Mining (ICDM) , pp. 3366–3370, 2016.
3- O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A.
Karpathy, C. V Jan, J. Krause, and S. Ma, “ImageNet Large Scale Visual Recognition
Challenge.”.

12 | P a g e
13 | P a g e

S-ar putea să vă placă și