Characterisation of Knock in Direct Injection Hydrogen Internal Combustion Engines

The internal combustion engine (ICE) has been the 'silver bullet' in powering machinery for the transportation, mining and construction industries. However, with existing and upcoming regulations on CO2 emissions, the industry is exploring the viability of fuelling ICEs with hydrogen as a carbon neutral alternative - notable examples include BMW, Toyota, Yamaha and JCB.

Current hydrogen combustion research focuses on achieving high brake thermal efficiency (greater or equal 45%) while keeping NOx emissions levels low by utilising direct injection fuelling strategies. This results in increased volumetric efficiency and allows for a more precise control of abnormal combustion events compared to port fuel injection. Nevertheless, topics such as combustion irregularities, turbocharger design for hydrogen-specific operation, heat transfer and injection strategy optimisation remain underresearched.

The main goal of this research project is to improve the state-of-the-art knock modelling for hydrogen ICEs in a 1D simulation environment - namely GT-Power, by creating computationally inexpensive knock models and validating them against experimental data. At present, there is a single fast-running commercially available knock model developed by GT, which is based on a neural network trained with hydrogen ignition delay times using the kinetics mechanism developed by Keromnes et al. (2013).

This model offers approx. 16 times quicker simulations compared to a H2/02/NOx kinetics mechanism. However, it suffers from several limitations - (1) it is based only on the chamber pressure, temperature and recirculated exhaust gases, but does not account for the presence of different knock-inducing species such as NOx molecules and radicals in the unburned zone of the working fluid; (2) It is based on a now-outdated kinetics mechanism, current state-of-the-art mechanisms include Konnov (2019), Polimi (2020), Kovacs et al. (2020) and Sun et al. (2022) ; (3) in comparison to previous knock models developed for gasoline engines, this model cannot be quantified using mathematical equations based on the Arrhenius equation.

Therefore, at present this project aims to deliver new knock models created using the current state-of-the-art kinetics mechanisms. These will be based on neural networks, similar to GT's proprietary knock model, but also accounting for the presence of knock-inducing nitrous oxides, as well as on the Arrhenius equation, which expands on previous gasoline-specific models and includes terms for the recirculated exhaust gas and NOx concentrations. The following actions have been identified:

1. Conduct a literature review on the topic of knock in ICEs with a focus on hydrogen ICEs and applications of kinetics mechanisms. 2. Assess the pathways to creating neural network-based models and Arrhenius equation-based models.

3. Develop neural network-based models and explore their predictive accuracy against simulations using the detailed kinetics mechanism and validate them using empirical data. 4. Develop Arrhenius equation-based knock models and compare against detailed chemical kinetics and experimental data.

The potential benefits of this project are significant improvements to the computational time of 1D ICE simulations, while retaining the knock prediction accuracy of the state-of-the-art knock models. This will ultimately lead to increased engine thermal efficiency and reduced calibration times, as the knock-limited spark advance can be better predicted prior to engine testing.