What is it about?

This work explores a new and simpler way to predict how fluids move when a surface is spinning, such as air or liquid flowing over a rotating disk. These kinds of flows appear in many everyday and engineering systems, including turbines, cooling devices, rotating machinery, and small aerospace components. Accurately predicting this motion is important for improving efficiency, stability, and performance, but traditional computer simulations can be slow, complex, and expensive because they require very fine computational grids and large computing power. In this study, machine learning models are combined with basic laws of physics to create a faster and more flexible prediction tool. Instead of learning only from data, the models are trained to obey the fundamental rules that govern fluid motion. This allows the method to predict flow behavior without relying on complicated meshes used in conventional simulations. The approach is tested on well-known spinning flow problems and compared with established analytical solutions and high-quality numerical simulations. The results show that physics-informed machine learning can reliably capture key flow features while reducing computational effort. Overall, this work highlights the potential of combining physics and artificial intelligence as a practical alternative for studying complex fluid flows in science and engineering.

Featured Image

Why is it important?

This work is important because it addresses a long-standing challenge in fluid mechanics: accurately and efficiently predicting complex spinning flows. Traditional simulation methods can be slow, computationally expensive, and difficult to apply to low-speed or near-wall flows, which are common in many modern engineering systems. As designs become smaller and more precise, these limitations become even more significant. What makes this work unique is the use of physics-informed machine learning, which combines artificial intelligence with the fundamental laws of fluid motion. By embedding physics directly into the learning process, the approach reduces the need for heavy computational meshes while still producing physically reliable results. The study also provides a clear comparison between different physics-informed models, highlighting their strengths and limitations for both steady and time-dependent flows. This research is timely because physics-informed ML is rapidly emerging as a powerful tool in science and engineering. The framework presented here demonstrates how such methods can complement or even replace traditional simulations, potentially saving time and computational resources. As a result, this work can help accelerate the design and analysis of rotating systems in aerospace, energy, and mechanical engineering, making advanced flow prediction more accessible to researchers and engineers.

Perspectives

From my perspective, this work represents an important step in learning how physics-informed machine learning can be applied to real fluid mechanics problems rather than idealized examples. Working on spinning flows highlighted both the strengths and the current limitations of these methods. While the models can accurately capture steady flow behavior with relatively low computational cost, they also reveal challenges in handling slow, viscous, and time-dependent flows at low Reynolds numbers. This study has shaped my understanding of how closely physics and data-driven methods must be integrated to produce meaningful results. It reinforced the idea that neural networks are not replacements for physical insight, but tools that work best when guided by well-established laws. Through this work, I also gained practical insight into model design, training stability, and the trade-offs between accuracy and efficiency. Looking ahead, I see strong potential for extending this approach to more complex geometries, higher Reynolds numbers, and coupled physical processes such as heat transfer. I hope this work encourages further exploration of physics-informed learning as a complementary tool to traditional simulations, especially for problems where conventional methods are costly or difficult to apply.

Kunal .
Indian Institute of Technology Kanpur

Read the Original

This page is a summary of: A Physics-Informed Machine Learning Approach for Modeling and Predicting Complex Spinning Flows, January 2025, American Institute of Aeronautics and Astronautics (AIAA),
DOI: 10.2514/6.2025-105504.
You can read the full text:

Read

Contributors

The following have contributed to this page