Session: 06-06 Abnormal Combustion Modeling

Paper Number: 191318

191318 - Machine Learning Enabled Robust Combustion Classification in a Hydrogen Engine 

Abstract: 

Hydrogen-fueled internal combustion engines are a promising decarbonization pathway for the transport sector. However, their performance is constrained by abnormal combustion phenomena, particularly pre-ignitions. The high frequency of pre-ignitions in hydrogen engines can render conventional heuristics based on cylinder-pressure-derived cyclic metrics unreliable for robust cycle classification and pre-ignition identification. Machine learning approaches that can learn distinguishing features of different cycle classes offer a promising alternative.

In this study, five macine learning based methodologies were evaluated to classify cyclic data from a heavy-duty, spark-ignition, port fuel injected hydrogen engine. Cycles were categorized into four classes: normal, weak pre-ignition, strong pre-ignition, and knocking. The evaluated approaches spanned unsupervised and supervised techniques, including both physics-agnostic and physics-informed frameworks. Models were trained using either windowed in-cylinder pressure traces or selected cyclic features (peak pressure and its location, knock intensity, start of combustion, ignition timing, and burn rate). Among the tested methods, the two physics-informed approaches, namely pre-stratified hierarchical agglomerative clustering and a dual-branch neural network, achieved the highest cycle classification and pre-ignition detection accuracy. The results highlight that incorporating physical insight into machine learning models improves their abnormal combustion detection performance. Such machine learning models can be used to develop suitable combustion control systems for hydrogen engines. 

Presenting Author: Abdullah Bajwa SwRI

Presenting Author Biography: Abdullah is a mechanical engineer who completed his graduate studies at Texas A&M University as a Fulbright scholar, where his research focused on gas exchange processes in natural gas two-stroke engines to reduce NOx emissions. He subsequently spent two years as a postdoctoral researcher at the University of Oxford, collaborating with Jaguar Land Rover on advanced gasoline combustion concepts and hydrogen internal combustion engine development. Since January 2024, he has been with Southwest Research Institute, supporting alternative fuel vehicle development as part of the Engine Systems Research & Innovation team.

Authors:

Jizhong He UTSA
Abdullah Bajwa SwRI
Yuanxiong Guo UTSA
Vickey Kalaskar SwRI
Ryan Williams SwRI