Bayesian Estimation of Transient Engine Exhaust Composition from Fourier Transform Infrared Spectroscopy Measurements
Date of Award
Doctor of Philosophy (PhD)
Mobile sources comprise a substantial portion of anthropogenic volatile organic compound (VOC) emissions worldwide. Many research efforts have sought to elucidate the relationships between VOC emissions and engine operating conditions, which are largely transient in real-world scenarios. However, the literature remains dominated by steady-state data and batch measurements of total emissions over entire driving cycles. Fourier transform infrared (FTIR) spectroscopy is a promising technique for obtaining instantaneous, time-resolved VOC measurements. However, FTIR measurements of chemically evolving samples are biased due to sample recirculation and signal non-stationarity. To extract accurate emissions profiles from biased FTIR measurements of transient emissions, an Unscented Kalman filter (UKF) is developed. The UKF is a model-based algorithm which incorporates sample mixing dynamics, a measurement model of non-stationarity effects, and noise statistics to infer instantaneous exhaust composition in a statistically optimal manner. The sample mixing model is developed and validated using computational fluid dynamics and mixing network simulations. Non-stationarity effects – which produce FTIR measurements that are unevenly weighted by spectral IR powers at/near the centerburst position of the modulating mirror – are mathematically and experimentally proven to emerge due to alternating scan directions. A numerical method is developed to estimate the degree of centerburst weighting on measurements of unique VOCs, which is mathematically shown to scale with spectral absorbance broadening. The UKF is experimentally validated by flowing transient, trace quantities of acetylene and propylene through a FTIR gas cell and filtering the resulting measurements. Average improvements of 58% and 51% are achieved for estimations of acetylene and propylene composition, respectively, compared to unfiltered FTIR measurements. The UKF is employed to investigate transient effects on emissions of various fuel component VOCs (cyclohexane, ethanol and pentane) and intermediates (1,3 butadiene, acetylene, ethylene, formaldehyde and methane) from a spark-ignited, port fuel-injected gasoline engine under various load ramps. Deterministic transient effects are evident, as emissions deviate from quasi-steady predictions by statistically significant quantities in 14 of the 21 species/load profile combinations explored. For the intermediate species, greater quasi-steady prediction errors correspond to faster ramp rates, while greater errors occur during moderate load ramps for fuel component species.