Aerospace Material Defect Detection

This project focuses on detecting surface defects in aerospace materials using machine learning and computer vision techniques.
The goal is to automate defect identification to improve inspection efficiency and aviation safety.

Problem Statement

Aircraft surface defects such as cracks, dents, and corrosion can compromise structural integrity.
Manual inspection is time-consuming and prone to human error, motivating the need for automated defect detection.

Project Objectives

• Detect surface defects in aerospace components
• Apply convolutional neural networks (CNNs) for image classification
• Evaluate model performance using standard metrics
• Demonstrate real-world applicability through deployment

Methodology

• Dataset preprocessing and image normalization
• Transfer learning using ResNet50
• Binary classification: defect vs no defect
• Model evaluation using accuracy, precision, recall, and F1-score

Results

• Training accuracy: ~92%
• Validation accuracy: ~81%
• Strong precision with reasonable recall
• Visual validation through sample predictions

Deployment

The trained model was deployed using Gradio and Hugging Face Spaces, allowing users to upload images and receive real-time predictions.

Tools & Technologies

• Python
• TensorFlow / Keras
• CNNs (ResNet50)
• Gradio
• Hugging Face Spaces

Academic Context

This project was completed as part of AI 2010 – Introduction to Applied AI at Iowa State University.

Author

Shreshta Anupindi
Aerospace Engineering Student
Iowa State University