Once trained, machine learning-based simulations can make these assessments almost instantly. Instant feedback enables rapid design testing, real-time adjustments, and rapid test changes without the usual computational burden.
The survey results are Journal of Computational Physics.
This research uses the stability of fluid motion as the basis for a new method to predict how complex systems will behave. Rather than relying on expensive laboratory experiments, solutions to the fundamental equations of fluid motion are generated numerically. This allows machine learning models to be trained on accurate, high-quality data extracted directly from physics, demonstrating that the models can accurately handle difficult simulations.
The main focus of this study is to identify bifurcation points, the moments when a smooth, steady flow (laminar) suddenly begins to change. This is similar to when a calm, evenly flowing river hits an obstacle or splits, causing the fluids to mix and begin to form eddies. Laminar flow is when a liquid behaves in a smooth, orderly manner, and the flow is consistent and steady, like pouring honey.
By successfully using a machine learning model to identify points, in this case bifurcations, at which a system’s behavior changes, this study suggests that with further refinement, machine learning-based models could become a practical alternative to traditional fluid modeling techniques in the future.
Professor Sylvester added: “This fusion of old and new approaches holds promise for efficiently calculating physically realistic fluid flows in a myriad of practical situations. In the future, the development of sophisticated mathematical models of complex fluids may be critical if the promise of AI is to be effectively realized.”
