search

SayanSomya / Defect-Detection-in-3D-Printing

Bachelors Thesis Project for Defect Detection
0 stars 0 forks source link

readme

Defect Detection in 3D Printing

Bachelor's Thesis Project

Welcome to the official repository for Defect Detection in 3D Printing, a Bachelor's Thesis project focused on leveraging advanced object detection models (YOLOv5 and YOLOv11) to identify defects in 3D-printed objects. This project combines computer vision, machine learning, and domain-specific problem-solving to enhance the quality control process in additive manufacturing.

Project Overview

Additive manufacturing, or 3D printing, has revolutionized how we create objects by enabling complex, highly customizable designs. However, the quality assurance of 3D-printed components remains a significant challenge due to defects that may occur during printing. This project addresses this challenge by implementing state-of-the-art object detection models to automate defect detection, thereby reducing the reliance on manual inspection.

Key Features:

  1. Automated Defect Detection: Identify common defects such as layer shifts, stringing, and under-extrusion.
  2. YOLO-based Object Detection Models: Evaluate and compare YOLOv5 and YOLOv11 for detecting anomalies in 3D prints.
  3. Custom Dataset: Built a comprehensive dataset containing annotated defect images for model training and testing.
  4. Performance Optimization: Focused on precision, recall, and inference time to ensure practical usability in real-time applications.

Methodology

  1. Dataset Preparation: Labeled the dataset with bounding boxes for precise defect localization & performed image augmentation to improve model generalization.
  2. Model Training: Implemented YOLOv5 and YOLOv11 in Google Collab and optimized hyperparameters for better accuracy and inference speed.
  3. Evaluation: Compared YOLOv5 and YOLOv11 based on precision, recall, F1-score, and mAP (Mean Average Precision).

Future Work

  1. Extend the dataset with more defect types and variations.
  2. Explore other state-of-the-art object detection architectures like YOLOv11 and DETR.
  3. Integrate the detection system into a 3D printing pipeline for real-time defect monitoring.