Data-driven prediction of spray macroscopic characteristics for marine injectors using neural networks

• Deep Neural Network (DNN) for predicting marine injectors spray penetration and cone angle; • Experimental data from high-speed imaging for the DNN training; • DNN exhibits strong predictive capabilities and reduced computational cost; • Exploration of feature importance and geometrical design space for target spray morphology. Fuel flexibility ensures reliable operation and improves the implementation of a dual-fuel strategy alongside the Diesel-only model in marine powertrains. This versatile approach, however, imposes limitations related to the complexity of the injection system, underpinning the necessity of comprehending the relationship between design and performance to facilitate the injector optimization process. The present study introduces a data-driven predictor utilizing a deep neural network to predict spray tip penetration and cone angle in marine injectors. This neural network is trained using experimental data from serial and prototype industrial designs, incorporating operating conditions and a wide range of geometrical parameters as primary input features. Issues, as for example overfitting of the training dataset, were mitigated via regularization, enhancing generalization. The deep neural network accurately predicts spray characteristics across short, medium, and long penetration ranges, achieving 95% accuracy for unseen data. Furthermore, a feature importance analysis indicates that the injection pressure, number of spray holes, outlet spray hole diameter, and sac hole volume are the primary parameters influencing spray behavior. This neural network provides a computationally efficient alternative to conventional approaches, such as time-consuming Computational Fluid Dynamics simulations or test measurements. The model is tailored to support the marine injector design workflow, allowing the fast exploration of design space in the early design phase at operation conditions relevant for fuel flexibility, and contributing to accelerate the injector development process.