Study of Impinging Liquid in Scramjet Engine using Deep Learning Techniques
Scramjets at Supersonic Speed suffers from inefficient mixing in Combustion-Chamber, but with optimal setup, mixing can be made efficient. Deep learning techniques were used to predict the best design conditions by adopting Variational Autoencoder method for generating new images with an extra Neural Net Block to generate corresponding design conditions. ...learn more
Project status: Published/In Market
Intel Technologies
Intel Opt ML/DL Framework
Overview / Usage
Human’s desire to travel in supersonic speed can be made possible only through Scramjets. But the small residence time available for mixing and combustion of fuel and air in Scramjets act as major stumbling block in the technological development. Numerous studies were done on gaseous fuel though they cause problem in storage and availability there is limited information in liquid fuelled Scramjets. Limitation of carrying out experiment and the high time consuming Large Eddy (LES) simulations reduced the number of Models that can be tested. This Project address these matters: (a) By adopting a backward faced injector issuing doublet impinging liquid parallel to the supersonic flow, (b) Predicting an optimal design using Machine learning.
Methodology / Approach
A total of 4 Network Architecture were developed (a) Encoder, (b) Decoder, (c) Variational Autoencoder (d) Neural Net (NN2). The architectures for Encoder & Decoder is inspired from Squeeze-Net and Inception-Net
Technologies Used
Computing Environments:
- MatLab
- Python
Frameworks/Libraries:
- Keras
- Tensorflow (Backend)
- Matplotlib, Numpy etc.
Experiments:
- Back light Shadowgraphy
- Mie Scattering
- PDPA
Computing:
- Intel Xeon, 2.10GHz x16
- Nvidia Quadro P2000 GPU
- Intel Cloud Cluster for Showcasing
Repository
https://github.com/matrixBT/Scramjet-Learning
Collaborators
There are no people to show.