Date of Award

Summer 2023

Access Type

Dissertation - Open Access

Degree Name

Doctor of Philosophy in Electrical Engineering & Computer Science


Electrical Engineering and Computer Science

Committee Chair

Radu F. Babiceanu

First Committee Member

Eduardo A. Rojas-Nastrucci

Second Committee Member

Laxima Niure Kandel

Third Committee Member

Kenji Yoshigoe

Fourth Committee Member

Prashant Shekhar

College Dean

James W. Gregory



The defect detection problem is of outmost importance in high-tech industries such as aerospace manufacturing and is widely employed using automated industrial quality control systems. In the aerospace manufacturing industry, composite materials are extensively applied as structural components in civilian and military aircraft. To ensure the quality of the product and high reliability, manual inspection and traditional automatic optical inspection have been employed to identify the defects throughout production and maintenance. These inspection techniques have several limitations such as tedious, time- consuming, inconsistent, subjective, labor intensive, expensive, etc. To make the operation effective and efficient, modern automated optical inspection needs to be preferred. In this dissertation work, automatic defect detection techniques are tested on three levels using a novel aerospace composite materials image dataset (ACMID). First, classical machine learning models, namely, Support Vector Machine and Random Forest, are employed for both datasets. Second, deep CNN-based models, such as improved ResNet50 and MobileNetV2 architectures are trained on ACMID datasets. Third, an efficient defect detection technique that combines the features of deep learning and classical machine learning model is proposed for ACMID dataset. To assess the aerospace composite components, all the models are trained and tested on ACMID datasets with distinct sizes. In addition, this work investigates the scenario when defective and non-defective samples are scarce and imbalanced. To overcome the problems of imbalanced and scarce datasets, oversampling techniques and data augmentation using improved deep convolutional generative adversarial networks (DCGAN) are considered. Furthermore, the proposed models are also validated using one of the benchmark steel surface defects (SSD) dataset.

Shashi_Bhushan_Jha_GS9.pdf (368 kB)
GS9 Form