Stable, high-temperature acoustic sonoluminescence using learning-based control

Abstract

Abstract Achieving high collapse temperatures under stable operating conditions is a central challenge in acoustic cavitation and sonoluminescence, as instabilities typically limit energy focusing. Conventional high-amplitude, low-frequency driving favors high collapse temperatures but simultaneously triggers shape instabilities that may destroy the bubble before extreme conditions are reached. Here, we explore autonomous learning-based control strategies to actively direct the dynamics of an acoustically driven cavitation bubble, with the dual objective of enhancing stability and maximizing collapse temperatures. An adaptive agent, trained within a reinforcement learning framework, interacts with a numerical bubble model by modulating the amplitude and frequency of the driving acoustic field on a cycle-by-cycle basis, enabling the systematic discovery of time-dependent excitation protocols beyond conventional approaches. The control strategy is trained on a reduced-order model combining the Keller-Miksis equation with an ideal gas state description, and subsequently evaluated, once a stable strategy was obtained, on a more comprehensive and experimentally validated cavitation framework incorporating additional physical effects. In contrast to standard stationary sinusoidal driving, which often leads to unstable dynamics and limited heating, the learned control policies identify simple, experimentally feasible excitation patterns that stabilize bubble dynamics while substantially increasing collapse temperatures. Performance improvements of up to 7- and 9.5-fold in the training and testing models, respectively, are observed, depending on the initial bubble radius. These results highlight the potential of autonomous, learning-based control as a general framework for cavitation management, providing a pathway to reliably access extreme conditions in sonoluminescence and enabling new opportunities in applications ranging from sonochemistry to biomedical technologies.