AIRCRAFT WING SHAPE OPTIMIZATION USING MACHINE LEARNING

The research project investigates the use of machine learning techniques, specifically Support Vector Machine (SVM), Random Forest (RF), Gradient Boost (GB), and K-Nearest Neighbor (KNN), for predicting aerodynamic coefficients of airfoils under varying conditions. Notably, RF and GB exhibit promising performance across different scenarios, including predicting coefficients for new airfoil configurations and scenarios with significant changes in Reynolds numbers. Additionally, the study introduces BézierGAN, a Convolutional Neural Network (CNN)-based algorithm, for generating aerodynamic shapes efficiently, thus reducing computational costs. Further, Model Order Reduction (MOR) techniques are applied to develop lower-order models of turbulent airflow around airfoils, effectively reducing computational expenses while maintaining essential characteristics. The research also explores the IRKA-based MOR technique, which minimizes computational barriers during computational fluid dynamics (CFD) simulations around airfoils, demonstrating significant reductions in computational time. Overall, these findings offer promising avenues for optimizing airfoil designs and improving their performance in real-world applications.

Mohammad Monir uddin Associate Professor, Department of Mathematics & Physics, North South University, Dhaka, Bangladesh
Mohammad Niaz Morshed Lecturer at North South University, Dhaka, Bangladesh
Kife Intasar Bin Iqbal PhD Student
Md. Tanzim Hossain Lab Instructor & Research Assistant (RA) at North South University, Dhaka, Bangladesh
Azizul Haque Student at Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany

DEEP LEARNING IN MEDICAL IMAGING

Within our deep learning research, medical imaging holds a prominent place as one of our key research topics. Specifically, the majority of our efforts are concentrated on segmentation tasks. We've tackled significant challenges in this domain, including:

  • Segmenting prostate anatomy and tumors
  • Identifying regions associated with multiple sclerosis
  • Segmenting the intricate Circle of Willis in brain images

Right now, we're delving into some cutting-edge techniques like Vision Transformer, Diffusion and Mamba, aiming to further refine our segmentation models and push the boundaries of accuracy and efficiency.

Practical Solution

  • Beyond research, we've also built a practical solution. We've developed a web application where clinicians can upload their 3D medical images and in return, they get precise segmentations for each slice of those images. This tool is a game-changer, saving clinicians valuable time and effort in manually labeling images.

Future Vision

  • Looking forward, our ultimate goal is to revolutionize the medical system, making it fully digitalized. Imagine a world where machines assist clinicians in labeling radiology scans, making the process much faster and more accurate. This digitization has the potential to significantly improve patient outcomes by streamlining diagnosis and treatment planning.