During stable cavitation, the maximum peak of pulsation, the first rebound peak, and the time interval between these two peaks are all important parameters affected by the bubble equilibrium radius, the driving amplitude, and the driving frequency. However, due to the complex dependence, it is highly non-trivial to express the nonlinear relationships among them via definite formulas. In the paper, the backpropagation neural network is proposed to deal with these mappings by self-training and learning the inherent rules within the dataset. The Keller–Miksis equation is used to provide training and testing datasets for a neural network. Meanwhile, an empirical method is introduced to guide the selection of neuron number in the hidden layer. After training, the predictions from the network are checked against the theoretical values calculated from the equation. The results exhibit that the neural network could achieve excellent predictions and has a good ability to generalize. This work shows that artificial intelligence could provide a feasible way for future research on cavitation bubbles.
Duo Tao (Wed,) studied this question.