UCLA

This dissertation details the development and application of mid-infrared laser absorption spectroscopy sensing methods towards advancing low-carbon reciprocating engines for high-efficiency and low-emission power generation in a decarbonized energy sector. The scope of this work includes advancement in methods for fundamental spectroscopic studies, integration of advanced sensors into production reciprocating engines for characterization of combustion of low-carbon fuel blends, and computational methods advancement for high-speed real-time signal processing. A high-temperature, high-pressure optical gas cell is designed to enable controlled studies of molecular absorption spectra at high temperatures (>1200 K) and high pressures (>200 atm) to validate spectroscopic parameters at the elevated conditions in combustion engines. A novel optical approach provides access to the mid-wave infrared wherein lies the fundamental rovibrational absorption bands of combustion species critical to characterization of combustion process and emissions formation. Laser absorption sensors are developed and utilized for experimental measurements in the exhaust of a production Honda single-cylinder spark-ignition engine through design of an in-line exhaust sensor module to gain optical access to exhaust gases close-coupled to the exhaust valve. High-temperature opto-mechanical design and laser fiber-coupling assist in achieving robust measurements of cycle-resolved temperature and species (CO and NO) concentration at a rate of 10 kHz. The exhaust sensor is demonstrated by capturing cycle-to-cycle and intra-cycle emissions dynamics and characterizing emissions response to low-carbon fuel blends incorporating natural gas, hydrogen, and ammonia. To enable real-time measurement output at 10 kHz, computational time of the sensor data processing is reduced to sub-ms scales through the use of machine learning algorithms on an embedded processing platform. Compact neural network and ridge regression models are developed to calculate species concentration and temperature directly from transmitted laser signals, removing the need for computationally-intensive nonlinear fitting methods. The machine learning algorithms are deployed to a field-programmable gate array (FPGA) for further acceleration. Hardware-in-the-loop demonstration yields computational time and latency below 100 µs to expand use of the 10 kHz exhaust sensor for real-time sensing applications. Complementary to the sensor development work, a time-resolved chemical-kinetic model is constructed within Cantera to evaluate reciprocating engine performance and emissions during fueling with low- and non-carbon blends. The simulation model provides insights into strategies for optimization of low-carbon combustion and serves as a foundation for sensor interpretation and future work in engine optimization. Discussion of ongoing work includes the design and development of an electro-hydraulic camless valvetrain for future integration into a reciprocating engine architecture to enhance adaptability for fuel-flexible operation.