Abstract:
Machine learning is quickly becoming a significant scientific computer tool, with enormous potential to broaden the field of computational fluid dynamics. Various research has been conducted in the recent past which emphasized on how the use of different Machine Learning algorithm is playing an important role in the enhancement of computational fluid dynamics. In this work we try to discuss about how we created an architecture of a Machine Learning algorithm by using U-net, which is a type of convolutional neural network and tried to apply it in order to reproduce a flow field around a 2D airfoil, which can be easily, if not quickly, produced using CFD analysis. The principal aim of this thesis is to check whether our Deep learning architecture is capable of providing an acceptable prediction of the flow field or not. If the flow field from DL matches with that of CFD, then we can use this observation for further study and if it does not match, there is still room for further correction. In order to execute the experiment, 3325 CFD simulations were carried out and the flow fields achieved from the simulation as the result of the experiment were separated into two groups.80% of the data was taken for training the DL algorithm and 20% were used for validation. After implementing the DL algorithm, some of the results were found to be almost similar to the results produced from the CFD simulation.
Description:
Supervised by
Dr. Md Rezwanul Karim,
(Co-Supervisor)
Mr. Tahsin Sejat Saniat
Lecturer,
Department of Mechanical and Production Engineering (MPE),
Islamic University of Technology (IUT),
Board Bazar, Gazipur-1704, Bangladesh.
This thesis is submitted in partial fulfillment of the requirements for the degree of Bachelor of Science in Mechanical and Production Engineering, 2022.