A Machine Learning Approach for Hydrogen Internal Combustion (H2ICE) Mixture Preparation

Abstract

<div class="section abstract"><div class="htmlview paragraph">The present work discusses the potential benefits of using computational fluid dynamics (CFD) simulation and artificial intelligence (AI) in the design and optimization of hydrogen internal combustion engines (H2ICEs). A Machine Learning (ML) model is developed and applied to the CFD simulation data to identify optimal injection system parameters on the Sandia H2ICE Engine to improve the mixing. This approach can aid in developing predictive ML models to guide the design of future H2ICEs. For the current engine configuration, it is observed that hydrogen (H2) gas injection contributes mixing of H2 with air. If the injector parameters are optimized, mixture preparation is better and eventually combustion. A base CFD model is validated from the Sandia H2ICE engine data against Particle Image Velocimetry (PIV) data for velocity and Planar Laser Induced Fluorescence (PLIF) data for H2 mass fraction.</div><div class="htmlview paragraph">A Design of Experiments (DoE) was derived for the injection system parameters to train the ML model using Latin Hypercube Sampling (LHS). An ensemble ML emulator uses the results from different ML algorithms to emulate the CFD simulation model trained with DoE data. Considering the injection parameters, a merit function is derived to optimize injection parameters using stochastic models like Genetic Algorithm (GA) to get the best mixture preparation. A significant advantage to the DoE-ML approach is the CFD cases can be run concurrently, shortening the wall clock time dramatically. A sequential CFD optimization can take months to complete, while a DoE-ML optimization can be completed in days using high core count HPC resources.</div><div class="htmlview paragraph">The integration of CFD simulation and AI techniques can provide a powerful tool for the design and optimization of H2ICEs, enabling engineers to improve engine performance and reduce emissions while minimizing the need for expensive and time-consuming experimental testing. Furthermore, these techniques can help in the rapid and cost-effective evaluation of new H2ICE designs, accelerating the development and adoption of this promising technology.</div></div>