Reproduce “Correlation between morphological evolution of splashing drop and exerted impact force revealed by interpretation of explainable artificial intelligence” on OpenResearch

About this paper

Jingzu Yee, Daichi Igarashi, Pradipto, Akinori Yamanaka, Yoshiyuki Tagawa

Researchers at Tokyo University of Agriculture and Technology utilized an Explainable AI (XAI) video classifier to analyze high-speed recordings of liquid drop impacts, establishing a direct correlation between the temporal evolution of splashing-specific morphological features and the transient normalized impact force. This method identified critical impact times where morphological differences are most pronounced and demonstrated that key splashing features become most distinguishable shortly after the peak impact force is exerted.

Key insights

• XAI, even with a relatively simple FNN architecture, can effectively bridge the gap between complex visual phenomena in fluid dynamics and their underlying physical origins, offering deeper insights beyond mere prediction.
• The temporal evolution of specific morphological features, such as the lifted lamella in splashing drops and the thicker lamella film in non-splashing drops, is directly interpretable from the XAI model's weights.
• Critical morphological distinctions for classifying splashing events are most pronounced and influential for the AI at specific early-to-mid impact stages (normalized times of approximately 0.24 and 0.38), which may differ from human visual perception.

Summary by alphaXiv.