Active Learning for Human-In-The-Loop Fluid Dynamic Measurements

Experimental aerodynamic measurements are crucial for the development of cars, planes, and various applications, including renewable energy and HVAC systems. However, the extensive setup time required for these experiments often limits their practical use. Our research introduces a novel approach that eliminates setup time, allowing a focus on the measurements while ensuring output quality. The operator only needs to outline the domain of interest, after which our algorithms ensure that measurement time is optimally utilized to sample the underlying flow field accurately.

Although sensing and learning algorithms are widely applied in many robotic fields, their use in fluid dynamics is limited due to the challenges of real-time data access, manipulating ongoing measurements, and the availability of necessary computing infrastructure. Additionally, the accessibility of traversing systems is constrained by their reach, articulation, stability, and programming interfaces. Our method leverages the human operator to move the probe, with an augmented reality system guiding them to measure precisely according to the algorithmically defined locations.

We will showcase the current system implementation, the gains and shortcomings of this method, and discuss potential future developments enabled by the new capabilities of consumer-grade head-mounted displays.