https://github.com/asem000/wiggly-net

CNN-LSTM based network to predict the oscillatory motion of silicon jet impinging sharp density interface
https://github.com/asem000/wiggly-net

Last synced: over 1 year ago
JSON representation

CNN-LSTM based network to predict the oscillatory motion of silicon jet impinging sharp density interface

• Host: GitHub
• URL: https://github.com/asem000/wiggly-net
• Owner: ASEM000
• License: mit
• Created: 2020-09-16T06:39:20.000Z (almost 6 years ago)
• Default Branch: master
• Last Pushed: 2020-11-23T02:47:29.000Z (over 5 years ago)
• Last Synced: 2025-01-29T15:13:17.932Z (over 1 year ago)
• Language: Jupyter Notebook
• Size: 1.36 MB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
[![DOI](https://zenodo.org/badge/295945576.svg)](https://zenodo.org/badge/latestdoi/295945576)

# Wiggly-NET

CNN-LSTM based network to predict the oscillatory motion of silicone jet impinging sharp density interface

Part of the WINTER/SPRING 2020 undergraduate research program (URP) @ KOREA ADVANCED INSTITUTE OF SCIENCE AND TECHNOLOGY (KAIST)

The prediction is generated by recursive input into the model

### loss plot

![Image](https://i.imgur.com/Z983VXD.png)

### Example 1

**Ground truth (top) , recurisvely predicted (bottom)**

![Image](https://i.imgur.com/ZSwdD3b.png)

### Example 2

**Ground truth (top) , recurisvely predicted (bottom)**

![Image](https://i.imgur.com/ii3xuhD.png)

### Example 3

**Ground truth (top) , recurisvely predicted (bottom)**

![Image](https://i.imgur.com/ZkMs6UB.png)

### this is a substep in the analysis of this phenonmenon

![Image](https://i.imgur.com/wWrdWEj.png)

### simulation of this phenomenon using Ansys fluent [youtube link](https://www.youtube.com/watch?v=hXc3pOpEeXc&feature=youtu.be)

## How to use

### 0. Convert your video data into numpy array

convert your video to numpy array in the shape of 4D array ( number of frames , row size , col size , channel )

### 1. Load the converted array using Wiggly UI notebook

### 2.Choose the ROI you want predict its frames

Larger dimensions require longer training time as well as larger model

![Image](https://i.imgur.com/iQRbc2i.png)

### 3.Convert video to training data 

Define 

1. l=2,    **sequence length**

2. df=6,   **frame step**

3. test_size= 0.05,   **Test size 5%**

4. verbose=True    **Print useful info**