Physics driven behavioural clustering of free-falling paper shapes

Area of Science:

• Physics
• Complex Systems
• Machine Learning

Background:

• Complex physical systems often display diverse discrete behaviors.
• System complexity hinders traditional modeling and analysis.
• Existing machine learning methods lack physical insight into system behaviors.

Purpose of the Study:

• To develop a novel method for extracting physically meaningful clusters of discrete behavior from limited experimental data.
• To provide insights into the underlying physics driving different system behaviors.
• To demonstrate the method's efficacy on a V-shaped falling paper system.

Main Methods:

• A new method is presented for extracting physically meaningful functions from experimental observations.
• These functions are used to cluster discrete behaviors and enhance system understanding.
• The approach was applied to a V-shaped falling paper system with four distinct modes.

Main Results:

• The method successfully identified a set of candidate functions distinguishing behaviors.
• Behavioral classification achieved an error rate of 2.04%.
• The extracted functions provided insights into the physical phenomena governing each behavior.

Conclusions:

• The novel method effectively extracts physically meaningful insights from limited data.
• This approach aids in understanding and classifying complex system behaviors.
• The V-shaped falling paper system serves as a successful demonstration case.