Understanding and preventing damage from cloud cavitation

Cloud cavitation occurs when large clouds of bubbles form in a fluid and then rapidly collapse in regions of higher pressure. In some systems, such as pumps and artificial heart valves, this phenomenon can lead to material loss, vibrations, and performance drop.

High-speed photography is one way to understand the dynamics of cloud cavitation, but image processing is tricky — it’s time-consuming and can be sensitive to the condition of the photograph, making it a qualitative measurement rather than a quantitative one.

Hatzissawidis et al. developed an automated approach to analyze high-speed cavitation photographs utilizing deep leaning.

“Instead of manually analyzing thousands of photographs or developing a new image-processing procedure each time, we wanted to provide a tool that automatically identifies cavitation and extracts quantitative information,” author Grigorios Hatzissawidis said.

The researchers trained a neural network model using high-speed photographs of cloud cavitation on a hydrofoil, a type of wing used to generate lift underwater. They tested it on another part of the dataset not used to train the model and found that it reliably identified cavitating regions in the photographs.

“We then applied the complete method to sequences of high-speed photographs and were able to automatically distinguish the attached cavity sheet from detached clouds and follow their evolution,” Hatzissawidis said. “This allowed us to extract quantities such as the sheet length, cloud size and position, and their spatial distribution.”

Hatzissawidis hopes their open-source repository including the trained model, source code, and datasets — linked in the paper — will be utilized by researchers working with high-speed photographs of cloud cavitation.

“We would be very happy to see the software being used, adapted to other experimental data, and further developed by the research community,” Hatzissawidis said.

Source: “Deep learning semantic segmentation for cloud cavitation image analysis,” by Grigorios Hatzissawidis, Marlene Leimeister, Tobias Meck, Jonas Bergner, Emily Henn, Lara Kerres, and Peter F. Pelz, Physics of Fluids (2026). The article can be accessed at https://doi.org/10.1063/5.0345365 .