Prediction of self-excited pulsation characteristics in cavitation jet using a physics-guided grey wolf optimizer-genetic algorithm-XGBoost model

Cavitation jets have considerable potential for erosion-based applications. Predicting self-excited pulsation characteristics is challenging because of the complex, nonlinear interactions among operating parameters. Traditional models often lack sufficient precision and generalization capability for accurate prediction. To address this, this paper proposes a Physics-Guided Grey Wolf Optimizer-Genetic Algorithm-XGBoost (PG-GWO-GA-XGBoost) prediction model. A hybrid sampling method combining equidistant and Latin hypercube sampling was employed to sample the jet parameters, with corresponding self-excited pulsation characteristic data obtained through numerical simulation. The cavitation number was incorporated as a physics-based input feature, and a GWO-GA hybrid algorithm was utilized to optimize the hyperparameters of XGBoost. Interpretability analysis of the model was subsequently conducted using the SHAP method. The results demonstrate that the GWO-GA hybrid algorithm effectively integrates the global exploration capability of the GA with the local exploitation ability of the GWO, thereby achieving faster convergence and higher accuracy. The PG-GWO-GA-XGBoost model achieved high predictive accuracy on the test dataset within the investigated design space. For pulse peak, amplitude, and frequency prediction, it achieved R2 values of 0.9998, 0.9961, and 0.9956, respectively. Furthermore, the model showed certain extrapolation capability at three selected operating points outside the original design space, correctly identifying the jet states with reported prediction errors within 13%. SHAP analysis identified injection pressure as the most influential feature for pulse peak and amplitude, and standoff distance as the most influential feature for pulse frequency. Based on the model, response surfaces correlating injection pressure, confining pressure, and self-excited pulsation characteristics can be constructed, clearly defining the transition boundary from a pulsed jet to a continuous jet. This allows for the determination of the minimum required pumping pressure for field operations. The prediction model established in this research offers a decision-making basis for the application of cavitation jet technology.