An Automatic Highly Dynamical Digital Twin Design with YOLOv8 for hydrodynamic studies on living animals

This paper presents an innovative method using the YOLOv8 algorithm to automate the creation of digital twins (DTs) replicating hydrodynamic behaviors of real organisms in liquid environments. The approach extracts features from video data, facilitating efficient DT generation. Addressing computational challenges in accurately simulating fish movements, including computing the spatial and temporal derivatives of its boundary and the distance function, our method offers insights into animals’ perception and use of hydrodynamic cues. Training YOLO on a custom dataset, processing predictions to ensure coherence between frames, integrating with computational fluid dynamics (CFD), and comparative analysis against ground truth simulations demonstrate the effectiveness of our automated digital twin creation. The paper also includes a sensitivity analysis to explore the impact of different customization aspects and aiming to provide insights into potential avenues for improvement.