Data-Driven Misfire Detection in Hydrogen Gen-sets using a Production Exhaust Pressure Sensor

Abstract

With the growing demand for climate-neutral powertrains, hydrogen combustion gen-sets are emerging as cleaner alternatives to diesel gen-sets. However, spark-ignited hydrogen engines are prone to misfires, impacting performance and engine lifespan. This study presents a novel approach for detecting misfires and identifying the misfiring cylinder using exhaust pressure signals from the production sensor, enabling a cost-effective, real-time diagnostic solution. Unlike complex feature extraction methods, the proposed approach is optimized for constant-speed gen-sets, ensuring computational efficiency and seamless integration within an Engine Management System. The technique utilizes exhaust pressure and crank angle signals to compute a tracking error feature—the squared deviation between the actual pressure signal and a reference signal. A common reference signal is modeled using normalized normal combustion exhaust pressure data from the training set and can be used for different loads. The method is validated at a 6° crank angle resolution in the hardware across multiple misfiring patterns, including single, continuous, and multiple cylinder misfire events, and the results demonstrated excellent performance under steady-state conditions. Finally, validation on the research engine demonstrated the method’s feasibility for real-time implementation.