What is it about?

Splashing occurs when a drop falls and impacts on a surface at a velocity high enough which causes the drop to break up and eject secondary droplets. By using high-speed camera, we collected images of splashing and nonsplashing drops. By using the collected images, we trained an AI to classify splashing and nonsplashing drops. After the AI learnt to classify images of splashing and nonsplashing drops accurately, we checked how does the AI perform the classification. As expected, the main characteristics that the AI checks is the breaking up of the drop's main body and the ejection of the secondary droplets. Remarkably, the AI also checks the contour of the drop’s main body. In other words, the contour is high, the AI will classify the drop as a splashing drop. The difference in the contour height between splashing and nonsplashing drops has not been reported prior to our study.

Featured Image

Why is it important?

Usually, an AI that can accurately classify images is quite complicated and it is almost impossible to understand how it works. However, in our study, we managed to train a simple AI to accurately classify image of splashing and nonsplashing drops and to understand the classification process of the AI. As a result, the AI helped us to discover a difference between splashing and nonsplashing drops that has not been reported before, that is the contour of a splashing drop is higher than that of a nonsplashing drop. Therefore, by understanding how an AI works, we can use it to observe physical phenomena and find new perspectives from it.

Perspectives

It is very exciting to work on this project. To observe a drop splash in slow motion is one of the most beautiful things I have ever seen. Also, to find out how an AI thinks is one of the coolest things that I have ever done.

Jingzu Yee
Tokyo Noko Daigaku

Read the Original

This page is a summary of: Image features of a splashing drop on a solid surface extracted using a feedforward neural network, Physics of Fluids, January 2022, American Institute of Physics, DOI: 10.1063/5.0077050.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page