AI-driven acoustic data integration model for heat flux prediction during pool boiling

Pool boiling is a complex heat transfer process. The estimation of heat flux in the system during the pool boiling process still poses a challenge. This research work presents a non-intrusive, data-driven, and machine learning model for predicting heat flux. This work analyses the heat flux, liquid acoustic pressure waves, and solid acoustic stress waves (Acoustic Emission) acquired from the experiments on two different boiling surfaces. By applying feature engineering, the imbalance between the liquid and solid acoustic data is resolved. The integrated boiling acoustic data is used to train an XGBoost ensemble regressor to predict the heat flux. The regressor is trained by carefully tuning the hyperparameters. The trained models are interpreted using SHAP values, demonstrating the necessity of boiling acoustic data integration. The trained model could predict the heat flux at the two different boiling surfaces with low root mean squared error (RMSE). The RMSE to the standard deviation (RSD) of the predicted results indicate that the performance of the model is categorised as “very good”. Future work will attempt to improve the RMSE and performance of the model by integrating advanced feature engineering.