Date of Award

4-2023

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering (MSEE)

Department

Electrical and Communication Engineering

First Advisor

Falah Awwad

Second Advisor

Sheharyar Malik

Abstract

The primary objective of this thesis is to develop innovative techniques for the inspection and maintenance of aircraft structures. We aim to streamline the entire process by utilizing images to detect potential defects in the aircraft body and comparing them to properly functioning images of the aircraft. This enables us to determine whether a specific section of the aircraft is faulty or not. We achieve this by employing image processing to train a model capable of identifying faulty images. The image processing methodology we use involves the use of images of both defective and operational parts of the aircraft's exterior. These images undergo a preprocessing phase that preserves valuable details. During the training period, a new image of the same section of the aircraft is used to validate the model. After processing, the algorithm grades the image as faulty or normal.

To facilitate our study, we rely on the Convolutional Neural Network (CNN) approach. This technique collects distinguishing features from a single patch created by the frame segmentation of a CNN kernel. Furthermore, we use various filters to process the images using the image processing toolbox available in Python. In our initial trials, we observed that the CNN model struggled with the overfitting of the faulty class. To address this, we applied image augmentation by converting a small dataset of 87 images to an augmented dataset of 4000 images. After passing the data through multiple convolutional layers and executing multiple epochs, our proposed model achieved an impressive training accuracy of 98.28%.

In addition, we designed a GUI-based interface that allows users to input an image and view the results in terms of faulty or normal. Finally, we propose that the application of this research in the field of robotics would be an ideal area for future work.

Arabic Abstract

اﻟﮭﺪف اﻷﺳﺎﺳﻲ ﻣﻦ ھﺬه اﻷطﺮوﺣﺔ ھﻮ ﺗﻄﻮﯾﺮ ﺗﻘﻨﯿﺎت ﻣﺒﺘﻜﺮة ﻟﻔﺤﺺ وﺻﯿﺎﻧﺔ ھﯿﺎﻛﻞ اﻟﻄﺎﺋﺮات، ﻧﮭﺪف إﻟﻰ ﺗﺒﺴﯿﻂ اﻟﻌﻤﻠﯿﺔ ﺑﺄﻛﻤﻠﮭﺎ ﻣﻦ ﺧﻼل اﺳﺘﺨﺪام اﻟﺼﻮر ﻻﻛﺘﺸﺎف اﻟﻌﯿﻮب اﻟﻤﺤﺘﻤﻠﺔ ﻓﻲ ھﯿﻜﻞ اﻟﻄﺎﺋﺮة وﻣﻘﺎرﻧﺘﮭﺎ ﺑﺼﻮر اﻟﻄﺎﺋﺮة اﻟﺘﻲ ﺗﻌﻤﻞ ﺑﺸﻜﻞ ﺻﺤﯿﺢ. ﯾﺘﯿﺢ ﻟﻨﺎ ذﻟﻚ ﺗﺤﺪﯾﺪ ﻣﺎ إذا ﻛﺎن ﺟﺰء ﻣﻌﯿﻦ ﻣﻦ اﻟﻄﺎﺋﺮة ﻣﻌﯿﺒًﺎ أم ﻻ، ﯾﺘﻢ ذﻟﻚ ﻣﻦ ﺧﻼل اﺳﺘﺨﺪام ﻣﻌﺎﻟﺠﺔ اﻟﺼﻮر ﻟﺘﺪرﯾﺐ ﻧﻤﻮذج ﻗﺎدر ﻋﻠﻰ ﺗﺤﺪﯾﺪ اﻟﺼﻮر اﻟﻤﻌﯿﺒﺔ. ﺗﺘﻀﻤﻦ ﻣﻨﮭﺠﯿﺔ ﻣﻌﺎﻟﺠﺔ اﻟﺼﻮر اﻟﺘﻲ ﻧﺴﺘﺨﺪﻣﮭﺎ اﺳﺘﺨﺪام اﻟﺼﻮر ﻟﻸﺟﺰاء اﻟﻤﻌﯿﺒﺔ واﻟﺘﺸﻐﯿﻠﯿﺔ ﻟﻠﺠﺰء اﻟﺨﺎرﺟﻲ ﻟﻠﻄﺎﺋﺮة، ﺗﺨﻀﻊ ھﺬه اﻟﺼﻮر ﻟﻤﺮﺣﻠﺔ ﻣﺎ ﻗﺒﻞ اﻟﻤﻌﺎﻟﺠﺔ اﻟﺘﻲ ﺗﺤﺎﻓﻆ ﻋﻠﻰ اﻟﺘﻔﺎﺻﯿﻞ اﻟﻘﯿﻤﺔ، ﯾﺘﻢ اﺳﺘﺨﺪام ﺻﻮرة ﺟﺪﯾﺪة ﻟﻨﻔﺲ اﻟﺠﺰء ﻣﻦ اﻟﻄﺎﺋﺮة ﻟﻠﺘﺤﻘﻖ ﻣﻦ ﺻﺤﺔ اﻟﻨﻤﻮذج وذﻟك ﺧﻼل ﻓﺗرة اﻟﺗﺟرﺑﺔ. ﺑﻌﺪ ﻋﻤﻠﯿﺔ اﻟﻤﻌﺎﻟﺠﺔ، ﺗﻘﻮم اﻟﺨﻮارزﻣﯿﺔ ﺑﺘﺼﻨﯿﻒ اﻟﺼﻮرة ﻋﻠﻰ أﻧﮭﺎ ﻣﻌﯿﺒﺔ أو ﻋﺎدﯾﺔ.
ﻧﻌﺘﻤﺪ ﻓﻲ دراﺳﺗﻧﺎ ھذه ﻋﻠﻰ اﺳﺘﺮاﺗﯿﺠﯿﺔ اﻟﺸﺒﻜﺔ اﻟﻌﺼﺒﯿﺔ اﻻﻟﺘﻔﺎﻓﯿﺔ (CNN) ﺗﻘﻮم ھﺬه اﻟﺘﻘﻨﯿﺔ ﺑﺠﻤﻊ اﻟﺳﻣﺎت اﻟﻤﻤﯿﺰة ﻣﻦ ﺟﺰء واﺣﺪ ﺗﻢ إﻧﺸﺎؤه ﺑﻮاﺳﻄﺔ ﺗﺠﺰﺋﺔ إطﺎر CNN وﺑﻨﺎءً ﻋﻠﻰ ذﻟﻚ، ﻧﺴﺘﺨﺪم اﻟﻌﺪﯾﺪ ﻣﻦ اﻟﻤﺮﺷﺤﺎت ﻟﻤﻌﺎﻟﺠﺔ اﻟﺼﻮر ﺑﺎﺳﺘﺨﺪام أدوات ﻣﻌﺎﻟﺠﺔ اﻟﺼﻮر اﻟﻤﺘﺎح ﻓﻲ Python وذﻟك ﻓﻲ ﺗﺠﺎرﺑﻨﺎ اﻷوﻟﯿﺔ، ﻛﻤﺎ أﻧﮫ ﻻﺣﻈﻨﺎ ﺑﺄن ﻧﻤﻮذج CNN ﻛﺎﻓﺢ اﻹﻓﺮاط اﻟﺬي ﻗﺪ ﯾﺤﺼﻞ ﻓﻲ ﺗﺠﮭﯿﺰ اﻟﻄﺒﻘﺔ اﻟﻤﻌﯿﺒﺔ وﻟﻤﻌﺎﻟﺠﺔ ھﺬا اﻷﻣﺮ، ﻗﻤﻨﺎ ﺑﺘﻄﺒﯿﻖ ﺗﻜﺒﯿﺮ اﻟﺼﻮرة ﻋﻦ طﺮﯾﻖ ﺗﺤﻮﯾﻞ ﻣﺠﻤﻮﻋﺔ ﺻﻐﯿﺮة ﺗﺗﻛون ﻣﻦ 87 ﺻﻮرة إﻟﻰ ﻣﺠﻤﻮﻋﺔ ﻣﻌﺰزة ﺗﺘﻜﻮن ﻣﻦ 4000 ﺻﻮرة وذﻟك ﺑﻌﺪ ﺗﻤﺮﯾﺮ اﻟﺒﯿﺎﻧﺎت ﻋﺒﺮ طﺒﻘﺎت اﻟﺘﻔﺎﻓﯿﺔ ﻣﺘﻌﺪدة وﺗﻨﻔﯿﺬ ﻋﮭﻮد ﻣﺘﻌﺪدة، ﺣﻘﻖ ﻧﻤﻮذﺟﻨﺎ اﻟﻤﻘﺘﺮح دﻗﺔ ﺗﺪرﯾﺐ ﻓﺮﯾﺪة ﺑﻠﻐﺖ 98.28% .

ﻛﻤﺎ أﻧﮫ، ﻗﻤﻨﺎ ﺑﺘﺼﻤﯿﻢ واﺟﮭﺔ ﻗﺎﺋﻤﺔ ﻋﻠﻰ واﺟﮭﺔ اﻟﻤﺴﺘﺨﺪم اﻟﺮﺳﻮﻣﯿﺔ (GUI) ﺗﺴﻤﺢ ﻟﻠﻤﺴﺘﺨﺪﻣﯿﻦ ﺑﺈدﺧﺎل ﺻﻮرة وﻋﺮض اﻟﻨﺘﺎﺋﺞ اﻟﺘﻲ ﺗﻮﺿﺢ اﻟﺼﻮرة اﻟﻤﻌﯿﺒﺔ أو اﻟﺼﺤﯿﺤﺔ. وﻓﻲ اﻟﺨﺘﺎم، ﻧﻘﺘﺮح أن ﯾﻜﻮن ﺗﻄﺒﯿﻖ ھﺬا اﻟﺒﺤﺚ ﻓﻲ ﻣﺠﺎل اﻟﺮوﺑﻮﺗﺎت ﻣﺠﺎﻻً ﻣﺜﺎﻟﯿًﺎ ﻟﻠﻌﻤﻞ ﻓﻲ اﻟﻤﺴﺘﻘﺒﻞ.

Download
COinS