Vortex detection for fluid dynamics data

• المؤلف الرئيسي: Lim, Sheng Jie
• مؤلفون آخرون: Ke Yiping, Kelly
• التنسيق: Final Year Project
• اللغة: English
• منشور في: Nanyang Technological University 2023
• الموضوعات: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• الوصول للمادة أونلاين: https://hdl.handle.net/10356/166242
• المؤسسة: Nanyang Technological University

Vortex detection can provide benefits in a wide range of fields, including aerospace engineering and meteorology. Vortices are swirling patterns of fluid, which can bring both positive impact in processes like mixings, and negative impacts such as affecting the performance and safety of aircrafts. Traditionally, vortex detection methods can be classified into global and local methods. Global methods are accurate, with the trade off of performance, while local methods are faster but lose global information and thus have poorer performances Recent advances in machine learning have led to the development of new vortex detection methods that can improve both performance and efficiency. These new vortex detection algorithms borrow model architectures in the field of image segmentation. For example, Vortex U-net borrows the U-net architecture. In this paper, we explore adapting the features from Deeplab, namely atrous convolutions and Atrous Spatial Pyramid Pooling, and propose a novel new architecture, and compare the performance and efficiency of our new model with existing models. We also explore and evaluate fully connected Conditional Random Field as a post processing layer in our segmentation model.