Exploration of the Hysteresis in Speciated Emissions during Transient

May 24, 2019 - Experiments are performed at two constant speed settings of 1500 and 2000 rpm. .... To investigate acceleration and short-term transien...
0 downloads 0 Views 5MB Size
Download PDF
Article pubs.acs.org/EF

Cite This: Energy Fuels 2019, 33, 5620−5631

Exploration of the Hysteresis in Speciated Emissions during Transient Gasoline Engine Combustion David Wilson, Dylan Lehmier, and Casey Allen* Department of Mechanical Engineering, Marquette University, Milwaukee, Wisconsin 53233, United States

Downloaded via UMEA UNIV on August 20, 2019 at 09:45:27 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

S Supporting Information *

ABSTRACT: Emissions of various fuel components (cyclohexane, ethanol, and pentane) and reaction intermediate species (acetylene, ethylene, formaldehyde, and methane) from a multicylinder, port-fuel-injected, spark-ignited gasoline engine undergoing transient loads are measured using Fourier transform infrared (FTIR) spectroscopy. The load profiles explored herein consist of positive and negative load ramps spanning brake mean effective pressures of 2−7 bar lasting 1, 2.5, and 5 s, as well as periodic load ramps of identical magnitudes and durations. Experiments are performed at two constant speed settings of 1500 and 2000 rpm. Fourier transform infrared spectroscopy measurements are processed with a recently developed unscented Kalman filter [Wilson, D.; et al. Energy Fuels 2017, 31, 11156−11168; Wilson, D.; et al. Energy Fuels 2018, 32, 11899−11912], which combats the biasing effects of sample recirculation and signal nonstationarity associated with transient FTIR measurements, to improve emission estimations. Emissions during load transients are compared to quasi-steady model predictions and estimated emission stochasticity. Overall, the data shows that transient effects (i.e., load ramp rate, speed/load history, nonstoichiometric equivalence ratio) substantially influence volatile organic compound (VOC) emissions in a deterministic manner, as quasi-steady prediction errors regularly exceed the combined effects of stochasticity and uncertainty. Negative load ramps (from 7 to 2 bar) result in the greatest quasi-steady prediction errors of all load profiles. For the periodic load ramps, the greatest quasi-steady prediction errors of the intermediate and fuel component emissions occur for 1 and 2.5 s load ramps, respectively. In both cases, these errors surpass the 95% confidence interval of statistical significance for each species except cyclohexane. Benzene and toluene emissions are unreported due to low quantities and excessive measurement noise, whereas 1,3-butadiene emissions show minimal relation to engine speed/load. The results of this work suggest that transient and historical effects must be taken into account when predicting VOC engine emissions and that the quasi-steady approach is insufficient. in an air-quality study of two regions in Los Angeles.18 Vehicle and marine-vessel-related sources contributed 40−54% of the VOC emissions in Hong Kong in an air-quality study that transpired in 2006−2007.19 Furthermore, exhaust and evaporative emissions from gasoline vehicles were found to comprise 52% of the VOC emissions in Beijing from ambient composition measurements conducted in August 2005.20 Thus, improvements in air quality are contingent on progress in engine efficiency and control strategies to reduce the emission of harmful VOCs. To achieve progress in these areas, the relationships between engine operation and the quantity and distribution of VOC emissions must be well understood. Numerous studies have explored the relationships between VOC emissions and engine speed/load. Wang et al. measured the comprehensive composition of exhaust gas collected from three light-duty gasoline vehicles operating under the Economic Commission for Europe (ECE) and extraurban driving cycle (EUDC) test cycles using gas chromatography− mass spectrometry.23 It was found that the VOC emissions under the ECE, an urban, low-speed driving cycle, were 16 times greater than the VOC emissions of the EUDC, a faster, suburban driving cycle. In addition, aromatics and alkanes were

1. INTRODUCTION Volatile organic compounds (VOCs) are a group of carbonous, gaseous compounds that participate in atmospheric photochemical reactions.1 VOC emissions substantially influence the radiative properties and quality of the atmosphere, most notably by reacting with nitrous oxides (NOx) to produce tropospheric ozone, a potent greenhouse gas.2 Atmospheric lifetimes of greenhouse gases are also increased in the presence of VOCs due to competition for oxidants.3 Furthermore, VOCs form secondary aerosols, which scatter sunlight and promote cloud formation.4,5 Individual VOCs have distinct impacts on the atmosphere depending on their lifetimes, reactivities, and reaction mechanisms.6,7 For example, numerous studies have found that aromatics and alkenes possess high ozone creation potential relative to other VOC groups.8−10 Certain VOCs are also highly toxic.11 For instance, exposure to the known carcinogens benzene and 1,3-butadiene is linked with elevated cancer risk.12−15 Thus, species distribution should be considered when addressing the impact of anthropogenic VOC emissions. Vehicles contributed over 23% of anthropogenic VOCs emitted in the US in 2014 according to the National Emissions Inventory.16 Even greater contributions from vehicles are typical in urban areas,17−21 where elevated levels of smog and poor air quality are common.22 For example, approximately 71 and 80% of VOC emissions were attributed to mobile sources © 2019 American Chemical Society

Received: February 1, 2019 Revised: April 19, 2019 Published: May 24, 2019 5620

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels

Figure 1. Engine test cell schematic.

the most emitted groups during ECE and EUDC tests, comprising 38 and 37% of the distribution, respectively. Higher levels of the carcinogens toluene and benzene were also detected in the EUDC tests. Nakashima et al. measured concentrations of 54 different VOCs from exhaust from three gasoline vehicles operating under nine different driving cycles using gas chromatography-flame ionization detection.24 The distributions of alkanes, alkenes, aromatics, and aldehydes vary according to the driving cycle. For “cold” driving cycles, the alkanes and alkenes are present in roughly equal quantities in exhaust samples, whereas alkanes are dominant in “hot” cycles. In a review of VOC emission studies in China, Hong-li et al. compared VOC distributions of emissions for vehicles operating at idle operation and a steady speed of 20 km/h.25 It was found that alkanes (66%) and alkenes (50%) were dominant during idle and steady operations, respectively. Ethylene and propene, among a few other VOCs, were present in high quantities in the idle tests, whereas VOC distribution was more uniform during steady operation. The studies presented above illustrate that species distribution is a strong function of engine operating conditions. This suggests that speed/load points can be strategically selected within engine control algorithms to achieve emissions with desired VOC distributions. However, the development of such an algorithm requires a direct input−output relationship between engine speed/load and emissions. The emissions data from the studies presented above are from batch measurements, which provide average emissions over an operational regime but provide no information on emissions at specific speed/load points. Furthermore, previous research has shown that instantaneous engine emissions are a function of speed/ load history in addition to instantaneous speed/load for realworld driving conditions,26,27 which are largely transient.28 For example, Hagena et al. studied NOx and particulate emissions from a diesel engine for tip-ins from 1 to 9 bar at varying ramp rates.27 Particulate emissions during an instantaneous load change reached values 10 times greater than that of a tip-in spanning 5 s. Furthermore, predictions from a quasi-steady engine map underpredicted the peak particulate and NOx

emissions by about an order of magnitude and 33%, respectively. Discrepancies between quasi-steady predictions and actual emissions on a time-resolved basis have been reported in other instances in the literature,29−32 illustrating the influence speed/load history on emissions. To address the lack of time-resolved, transient VOC emission data in the literature and elucidate the relationships between species distribution of emissions and engine operational points, exhaust composition from a spark-ignited, portfuel-injected gasoline engine under various load profiles is measured using Fourier transform infrared (FTIR) spectroscopy. Positive and negative ramp load profiles with brake mean effective pressures (BMEPs) ranging from 2 to 7 bar at constant speed settings of 1500 and 2000 rpm are explored, which are typical conditions during low/partial load urban driving. To investigate acceleration and short-term transient effects on VOC emissions, ramp durations are varied between 1 and 5 s. Periodic ramp waves lasting up 30 s are also investigated to explore long-term historical effects. Emissions from these experiments are compared to predicted emissions from a quasi-steady model, which utilizes an emission map generated from measurements conducted at a range of constant engine speed and load points. The model predicts VOC emissions according to the map value that corresponds to the current engine speed and load. To address sample recirculation and spectral IR intensity stationarity issues associated with FTIR measurements of the sample with rapidly evolving composition, a previously developed unscented Kalman filter (UKF)33,34 is employed to filter FTIR measurements. This algorithm utilizes simple models for residence time distribution and sample absorbance evolution to estimate the composition of the sample entering the FTIR gas cell during a measurement period, thereby enabling timeresolved estimates of comprehensive emissions.

2. METHODS 2.1. Experimental Setup. Experiments are conducted on a 1.5 L Kubota WG1605 four-cylinder, port-fuel-injected, spark-ignited engine. This engine provides a maximum power output of 55 hp and operates at speeds between 750 and 3200 rpm and brake torques 5621

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels between 0 and 120 ft-lb. A 100 hp alternating current dynamometer from Power Test Inc. is coupled to the engine shaft to impart resistive loading. Engine control and data acquisition are accomplished using a combination of the National Instruments Engine Control System (ECS) and the PowerNet LT package (Power Test). The ECS is a LabVIEW-based control package that continuously passes speed and torque setpoints to PowerNet, which exercises ultimate authority over the engine and dynamometer. The operational parameters within the engine are controlled by a stock engine control unit provided by Kubota. For the speeds/loads explored in this work, spark timing is maintained between 15 and 21° BTDC, whereas the fueling pulse width ranges from 5 to 12 ms. A Bosch LSU 4.9 exhaust wideband oxygen sensor is employed to measure the air-to-fuel equivalence ratio (λ) from the exhaust line. Equivalence ratio measurements are reported at intervals of approximately 50 ms. Emission measurements are performed using an FTIR spectrometer. An FTIR determines the chemical composition of exhaust gas by emitting a modulated, broadband IR beam through an exhaust sample and measuring the incident IR intensity. The IR intensity (interferogram) is Fourier transformed with respect to the modulation profile and subtracted from a background spectrum, ultimately yielding an absorption spectrum that is indicative of the sample composition. FTIR spectroscopy is advantageous for this work because it measures many distinct VOCs simultaneously from continuously replenished exhaust samples, allowing VOC distribution to be directly mapped with engine speed/load. The FTIR utilized in this study is MKS 2030HS, which measures at a high frequency of 5 Hz, enabling transient emission profiles to be captured with the aid of the UKF. It has a spectral resolution of 0.5 cm−1, which permits absorption lines within close proximity to be resolved. This allows the presence of distinct species with similar absorption regions to be properly differentiated. Low detectability limits are attained with a long IR path length of 5.11 m. MKS 2030-HS is also equipped with a silicon carbide radiation source at 1200 °C, a liquid-nitrogen-cooled mercury cadmium telluride detector, and a helium neon laser used to generate a reference interferogram. Exhaust gas is sampled at a flow rate of 25 slpm. A layout of the experimental setup is provided in Figure 1. 2.2. Unscented Kalman Filter for FTIR Measurements. FTIR spectroscopy is an attractive diagnostic for comprehensive speciation due to its ability to distinguish individual VOCs according to their IR absorption spectra. However, FTIRs possess some characteristics that complicate measurements of chemically evolving samples. First, FTIR gas cells must have sufficient volume to allow multiple passes of the IR beam through the sample to achieve reasonable detectability limits. Because of its large volume, residence times within a gas cell can be greater than the measurement period of FTIR. Consequently, an FTIR gas cell contains exhaust gas from past engine cycles, resulting in biased FTIR measurements that are influenced by historical emissions. Furthermore, composition transients during an FTIR scan complicate the interpretation of the resulting IR absorption spectrum. For an interferogram from a chemically evolving sample, the calculated IR intensity spectrum by the Fourier transform is convoluted with the true spectral intensity profile with respect to IR modulation. This renders calibrations (which map calculated absorption spectra to steady compositions) insufficient for inferring the sample composition from FTIR measurements, and the profile of composition evolution with respect to modulation must also be considered. This issue has been discussed extensively in a previous work.34 The limitations of FTIR spectroscopy regarding transient analysis are addressed with the implementation of a previously developed UKF, which has been experimentally validated against numerous measurements of compositionally evolving samples. This algorithm and its validation have been thoroughly detailed in previous works33−35 and are therefore only briefly explained here. A UKF is an algorithm that couples model predictions with measurements to obtain statistically optimized estimations of the states of a system. The ultimate purpose of the UKF herein is to optimally estimate the composition of the sample entering the FTIR gas cell during a measurement period. Inlet composition is a more accurate

representation of instantaneous exhaust composition compared to gas cell composition, where samples from past engine cycles are present due to recirculation. The UKF employs a measurement model to infer the true gas cell composition from biased FTIR measurements of chemically evolving samples. FTIR measurements, as discussed extensively in ref 34, weigh absorbances (thus, compositions) near the beginning of a forward moving mirror scan (centerburst) higher than absorbances at subsequent times. Since moving mirrors alternate scan directions every measurement, this centerburst weighting creates artificial oscillations in unfiltered emissions data. The degree of centerburst weighting for each species is calculated by numerically convolving its calibration spectrum with a linearly evolving power profile, assuming that absorbance (or composition) evolves linearly with respect to modulation. Values for centerburst weighting for each species are provided in Table S1 in the Supporting Information. Calculated centerburst weighting is used within the measurement model to deduce the true gas cell composition that corresponds to a biased FTIR measurement. A simple residence time distribution model is used to infer the inlet composition from the estimated gas cell composition. An illustration of the estimation process of the UKF is provided in Figure 2. To statistically optimize the estimations,

Figure 2. UKF flowchart for optimally estimating engine exhaust composition from FTIR measurements. distributions must be known/assumed for measurement noise and state value fluctuations. The variance of the emission fluctuations for each species is estimated using the squared difference between the maximum and minimum emission values within the steady-state map divided by 4. For species with this value less than 5 ppm, the emission fluctuation is set to 5 ppm. Estimating the standard deviations of emission fluctuations in this way provides appropriate scaling between species, as species whose emissions cover greater ranges on the steady-state map are expected to have greater fluctuations during transients. The emission fluctuation standard deviation for each 5622

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels

Figure 3. Illustration of the process for estimating the magnitudes of FTIR noise effects on measured species. species is listed in Table S1 in the Supporting Information. The method for estimating the variance in composition measurements due to noise is slightly more involved and is detailed in the following subsection. 2.2.1. Influence of Noise on Measured Composition. For the UKF to accurately deduce the composition of engine emissions from FTIR measurements, the measurement noise statistics must be reasonably estimated. Since the UKF is expressed in terms of mass/molar compositions, it is advantageous to express the noise in terms of composition (as opposed to the signal level) as well. However, typical FTIR algorithmswhich convert the measured spectral intensity to chemical compositionuse the intensity at several wavenumbers to infer composition to reduce noise effects. Moreover, for a given level of signal noise, the corresponding composition noise becomes greater as absorbance (composition) increases due to the decreasing signalto-noise ratio. Even more, two or more species may absorb in the same spectral regions, and FTIR algorithms must be somewhat complex to account for this cross-talk. Thus, analytically determining the statistics of composition measurement noise is a difficult task and is not as simple as using the signal-to-noise ratio at a single wavenumber. Therefore, composition measurement noise is estimated in this work using a Monte Carlo-like method. The first step in this method is to select one or more measurements for which the noise statistics are of interest. For this work, a measurement at approximately the midpoint of the engine load transient for the 1500 rpm, 2.5 s positive load ramp is selected. Due to the tediousness of this method, it is only performed for this single measurement, and the resulting statistics are applied to all remaining measurements. This measurement is selected since it occurs at the midway point of the loads explored in this work and is from an experiment with a moderate ramp rate. In the second step, a series of FTIR noise measurements are conducted. In these measurements, the FTIR gas cell is filled with an unabsorbing gas (such as nitrogen) so that differences in spectral transmittance between each measurement are due to noise only. Thus, the spectral transmittance noise for a single measurement can be estimated by subtracting the average spectral transmittance from all measurements with that of the single measurement. Detector noise is the dominant form of noise in most commercial FTIRs.36 Since detector noise is additive,36 the effects of measurement noise can be simulated by adding each transmittance noise spectrum to the transmittance spectrum from the measurement of interest. The resulting composition noise variance for each species is then estimated by operating on each synthetic spectrum with the FTIR algorithm to calculate a corresponding composition. The variance of the composition distribution is then calculated. A diagram of this process for estimating composition measurement noise is provided in Figure 3. For this work, approximately 1500 noise spectra are utilized. This number of spectra is determined to be sufficient, as analysis with 750 spectra changes the composition variance by less than 5% for each species. Measurement noise variance for each species is provided in Table S1 in the Supporting Information. 2.3. Engine Load Profiles. Emission measurements from the Kubota engine are conducted for two sets of engine load profiles. The first set consists of positive and negative load ramps between 2 and 7 bar (in BMEP) with durations of 1, 2.5, and 5 s. Different ramp rates are explored to investigate how acceleration and short-term speed/

load history influence emissions. To examine the impact of long-term historical effects on emissions, a second set of periodic load ramps is imposed on the engine. The load in these profiles oscillates between the same extrema at the same durations as the single load ramps. These load oscillations are executed for approximately 30 s to allow possible long-term emission trends to develop. The initial load point is maintained for approximately 2 min before the first load change to eliminate prior transient effects on the emissions. Experiments are conducted at two constant engine speed settings: 1500 and 2000 rpm. For brevity, only the results for engine speeds of 1500 rpm are presented herein. The 2000 rpm results are provided in the Supporting Information. Two trials are performed for each load profile/engine speed combination to ensure repeatability. The transient load profiles are illustrated in Figure 4. Solid and dashed

Figure 4. Illustration of the transient engine load profiles. lines in Figure 4a represent the positive and negative load profiles, respectively. Blue, green, and red lines represent the 1, 2.5, and 5 s load ramps, respectively. These color distinctions for each ramp rate are maintained throughout the manuscript. To quantify the transient and historical effects on emissions, each emission measurement is compared with an estimation from a quasisteady model. The model calculates each quasi-steady estimation using an emission map generated from experiments at constant engine speeds and loads. These speed/load points are illustrated in Figure S1 in the Supporting Information document. The speed points range from 1000 to 2500 rpm with an interval of 250 rpm, whereas load (in BMEP) ranges from 2 to 7 bar with an interval of 1 bar. Measurements are recorded for a total of 3 min at each speed/load point to determine the corresponding mean emission value. Linear interpolation is used to estimate the emissions at intermediate speed/ load points. To determine if discrepancies between quasi-steady predictions and experimental emissions are due to transient effects (load ramp rate, speed/load history, rich/lean equivalence ratios) as 5623

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels

Figure 5. Emissions of various VOCs and engine speed/load for the negative load ramp, 1500 rpm cases. Error bars reflect the standard experimental emission uncertainty, whereas the shaded area surrounding the quasi-steady predictions represents the estimated standard emission stochasticity for the current engine speed/load. opposed to stochastic fluctuations, the standard deviation of emissions during each steady-state test is calculated for each speed/load point within the steady-state map. These calculations serve as an approximation for the stochasticity standard deviation at each speed/load point and are used to generate a similar map for stochasticity. Illustrations of the mean and standard stochasticity emission maps for each species are provided in Figures S2 and S3, respectively, in the Supporting Information document. It should be noted that “stochasticity” in this work refers to emission changes/ fluctuations that are not induced by transient effects. To deconvolve measurement noise effects from the calculated stochasticity, the steady-state measurements are filtered with the UKF before standard deviations are calculated. It should be acknowledged that transient effects may influence stochasticity. Thus, second trials are also conducted to determine whether a given emission fluctuation is mostly stochastic, whereas stochasticity calculations serve as first approximations.

determine if the differences between estimated experimental transient emissions and quasi-steady predictions are statistically significant, the UKF estimations are presented with error bars, which represent emission uncertainty to one standard deviation. For clarity, uncertainty bars are only present for every other estimation. Quasi-steady predictions are also presented with a surrounding shaded region that represents the predicted standard deviation of stochastic emission fluctuations. Such information provides an idea of whether the differences between the quasi-steady predictions and estimated emissions are simply due to stochastic fluctuations or if there is a deterministic component of the transient emissions that can be predicted with a more sophisticated model. The measured air-to-fuel equivalence ratio is presented at the end of the section since it likely impacts transient emissions and contributes to quasi-steady prediction errors. The equivalence ratio during steady operation is wellcontrolled and nearly stoichiometric. However, stoichiometry is more difficult to maintain while engine speed/load is changing. A series of three fuel components and four reaction intermediates are plotted for each transient profile. The fuel components are cyclohexane, ethanol, and pentane, whereas the intermediates are acetylene, ethylene, formaldehyde, and

3. RESULTS AND DISCUSSION The emissions of various species with respect to engine speed, load, and equivalence ratio for each transient profile are presented in this section. Transient engine emissions are estimated from FTIR measurements using the UKF. Predicted emissions from the quasi-steady model are presented alongside the estimated transient emissions for comparison. To 5624

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels

Figure 6. Emissions of various VOCs and engine speed/load for the positive load ramp, 1500 rpm cases. Error bars reflect the standard experimental emission uncertainty, whereas the shaded area surrounding the quasi-steady predictions represents the estimated standard emission stochasticity for the current engine speed/load.

methane. Due to the low concentrations and high measurement noise of benzene and toluene in this work, the ability to form substantial conclusions regarding the transient emissions of these species is limited. Thus, results for these species are not presented, but a discussion of the factors that contribute to their elevated levels of measurement noise is provided in the Supporting Information. Emissions of 1,3-butadiene are presented and discussed separately from the other species as it is the only species where its emissions appear mostly dominated by stochasticity. Most of the emission trends during each load profile are shared between the 1500 and 2000 rpm cases. For example, negative load ramps result in dramatic drops in fuel components and formaldehyde emissions, and advanced/delayed acetylene/ethylene increases, respectively, at both speeds. Quasi-steady predictions improve for positive load ramps for both speeds as well. The only notable differences are in formaldehyde and fuel component emissions during positive load ramps, where the former displays a complicated relationship with load and the latter exhibits a temporary upward spike during 1 s ramps for the 2000 rpm case. Due to these similarities, only the 1500 rpm cases are presented for brevity. Emissions for the 2000 rpm cases are

included in Section A.2 of the Supporting Information document. 3.1. Single Ramp Profiles. Figure 5 displays the emission profiles of each species for the negative load ramp cases, along with engine speed/load. Transient effects are apparent, as experimental emissions vary significantly from quasi-steady predictions in most cases, although the degree of deviation differs according to load ramp rate and species. The emission profiles for the fuel components (cyclohexane, ethanol, and pentane)which are displayed on the right-hand side of the figureare qualitatively similar in each case. For each of these species, the emissions rapidly drop at the onset of the load ramp, as opposed to a gradual rise as the quasi-steady model predicts. The two profiles are significantly separated at the trough of the drop, as the standard uncertainty bars of the experimental emissions lie well outside of the shaded area representing the standard stochastic deviations from the quasisteady predictions. Furthermore, while the experimental emissions eventually settle upward near the quasi-steady predictions, the settling time is delayed by roughly a second after the load ramp has concluded. Thus, fuel component VOC emissions are substantially influenced by transient effects/ historical engine conditions during negative load ramps, and 5625

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels

Figure 7. Emissions of various VOCs and engine speed/load for the periodic load profile, 1500 rpm cases.

The load ramp results in decreased emissions, whereas the quasi-steady model predicts an increase. Interestingly, ethylene emissions do not immediately respond to the load ramp but remain relatively steady for about a second after the initiation of the ramp. It should be noted that ethane emissions (which are omitted for brevity) also display a similar lagging effect with respect to negative load ramps. Formaldehyde arguably displays the most interesting emission trends of all of the species. For the 1 s ramp case, the rise predicted by the quasisteady model is equally matched by a rapid descent in the experimental emissions of >10 ppm. Furthermore, the experimental emissions fail to settle until approximately 5 s after the load ramp has completed. For the 2.5 and 5 s load ramps, the experimental emissions gradually descend by ∼5 ppm before rapidly ascending to their final, steady-state values at approximately 5 and 6 s, respectively. Both of these rapid ascensions occur after their respective load ramps are finished, suggesting a historical influence on formaldehyde emissions. Overall, these results indicate that intermediate species are affected uniquely by negative load transients, and the magnitude of these effects is dependent on the ramp rate.

the degree of this influence is similar regardless of species and the ramp rate. Intermediate species, on the other hand, each respond uniquely during negative load ramps, as shown on the left-hand side of Figure 5. For acetylene and methane, a 1 s ramp causes a significant upward spike in emissions that exceeds the magnitude predicted by the quasi-steady model by a factor of about 2. For the 2.5 and 5 s load ramp cases, a brief uptick in the emissions of acetylene and methane occurs after the conclusion of each ramp. The upticks appear statistically significant for both species, as the standard uncertainty bars of the experimental emissions lie outside of the shaded region surrounding the quasi-steady predictions. These upticks indicate a historical influence on emissions of these two species since they occur after the end of the load ramp. Methane emissions are also heavily influenced by stochastic and/or unrepeatable effects, as indicated by the wide shaded region of the quasi-steady predictions and rapid oscillations in experimental emissions, which appear uncorrelated to engine load. Experimental ethylene and formaldehyde emissions behave in an opposite manner as quasi-steady predictions. 5626

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels

profile. However, a deterministic positive spike in methane emissions can be discerned near the end of the 1 s load ramp. Furthermore, a significant portion of the emissions following the load ramp oscillates below the shaded stochasticity region surrounding the quasi-steady predictions, indicating a historical influence. For the other two load ramps, the experimental emissions appear mostly within the predicted regions by the quasi-steady model. As will be discussed more thoroughly in Sections 3.2 and 3.4, methane emissions may be sensitive to excursions from the stoichiometric air-to-fuel ratio, which are greater during faster load ramps. Thus, these equivalence ratio excursions may be partially responsible for the quasi-steady model errors that occur during the 1 s positive load ramp. 3.2. Periodic Ramp Profiles. To determine if there are any long-term relationships between emissions and engine load that are not discernible from single load ramp experiments, experiments with periodic load ramps are conducted. The resulting experimental emissions of both the fuel component and intermediate species are presented in Figure 7. For clarity, standard uncertainty bars and corresponding quasi-steady predictions are removed. Quasi-steady model errors are quantified later on in Figure 8. As shown in Figure 7, no

Figure 6 shows emissions of each species during the positive load ramps, along with engine speed/load. Generally, the discrepancies between the quasi-steady predictions and experimental emissions are modest compared with those of the negative load ramp cases, although significant disparity persists in some instances. Again, fuel component emissions are highly correlated between species. The quasi-steady model underpredicts the magnitudes of the cyclohexane and pentane spikes. Interestingly, experimental cyclohexane and pentane emissions also substantially lag their respective quasi-steady predictions for the 2.5 s ramp case; yet virtually no lag exists for the 5 s ramp case. This indicates a complex interplay between current and historical engine speed/load and that a sophisticated model may be needed to accurately predict the emissions of these species. Experimental ethanol emissions generally agree reasonably well with quasi-steady predictions during positive load ramps. The most glaring discrepancy for fuel component emissions occurs after the load ramps, as the experimental emissions settle at a greater composition than the quasi-steady prediction. This is likely not an error in the steady-state map. Recall that the final setpoint for the positive load ramps is the same as the initial setpoint for the negative ramps. In the former cases, initial fuel component emissions agree well with the quasi-steady predictions. The equivalence ratio may be responsible for these disparities, as will be discussed in Section 3.4, although it is interesting to note that this trend is not observed in the emissions of intermediates. Shown on the left-hand side of the figure are emissions of intermediate species during the positive load ramps. As in the negative load ramp cases, the profile for each intermediate species is relatively unique. For acetylene, the quasi-steady predictions agree reasonably well with the experimental emissions. The only glaring discrepancy occurs for the 1 s ramp case, where the experimental emission profile features an upward spike of approximately 7 ppm that begins nearly halfway through the load ramp. This spike is absent in the quasi-steady predictions. Reasonable agreement also exists for ethylene emissions. Each experimental emission point and/or its standard uncertainty lies within the shaded region of the quasi-steady predictions, aside from the troughs of the 1 and 2.5 s load ramps. However, even the disparities of these features are slight. These results indicate that the quasi-steady model performs adequately for acetylene and ethylene during positive load ramps. Formaldehyde emission profiles, however, consist of numerous oscillations that the quasi-steady model fails to predict. These include the initial positive and negative spikes at approximately 1 and 1.75 s during the 1 s load ramp, respectively, and the positive spike of ∼10 ppm midway through the 5 s load ramp. It is difficult to discern an obvious relationship between formaldehyde emissions and engine operation from this data, as emission oscillations do not immediately appear to correspond to changes in load or speed in a predictable way. However, these complex emission patterns are statistically significant and repeatable, indicating that they are indeed deterministic. Thus, formaldehyde emissions are influenced by transient operation in a complicated manner. Repeatability for formaldehyde emissions is demonstrated in Section A.3 of the Supporting Information document. Again, experimental methane emissions are mostly dominated by rapid fluctuations that are uncorrelated with engine load, as the shaded stochasticity region surrounding the quasisteady predictions spans ∼10 ppm throughout each load

Figure 8. Standard error between experimental emissions and quasisteady predictions for various species, compared with estimated standard emission stochasticity and uncertainty for the periodic load profile, 1500 rpm cases. Calculated reduced χ2 statistics lie above each bar, indicating the statistical significances of quasi-steady prediction errors relative to stochasticity and uncertainty.

long-term trends are easily discernible in the emissions for any fuel component or intermediate species, as the emission trends appear relatively stationary throughout each cycle. This suggests that historical effects on emissions, while significant as evidenced from the single load ramp experiments, do not extend beyond several seconds. This is important information for future emission models as it suggests an approximate limit on the amount of historical data needed to properly train or develop such models. For the most part, emissions of fuel components during the periodic load ramps behave as expected given data from the single load ramps. During the 2.5 and 5 s ramp waves, the fuel component emissions display an interesting pattern where a peak is reached midway through the positive load ramp, followed by a steady, modest drop that is capped off with an upward inflection at the end of the ramp. Furthermore, as the emissions begin to drop during the negative load ramp, the trough is reached well before the end of the ramp. In the context of the results for the single positive and negative load 5627

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels

indeed influence emissions in a compelling way. To rigorously estimate statistical significance of these errors, reduced χ2 statistics are also calculated for each species/profile combination and are presented above their corresponding bars in Figure 8. Higher χ2 statistics corresponds to higher likelihood that discrepancies between experimental and predicted emissions are due to deterministic, transient effects, as opposed to noise and/or random fluctuations. χ2 values are greater than 3.84 (which corresponds to 95% confidence) in 14 of the 21 cases, with the exceptions being all of the cyclohexane cases, ethanol at 1 s load ramps, acetylene at 2.5 and 5 s load ramps, and ethylene at 5 s load ramps. Cyclohexane emissions are confined within single digits in every load profile. At such small quantities, deterministic fluctuations are expected to rival uncertainty. The remaining exceptions are consistent with the overall tendencies observed in the data, which are discussed below. For the intermediate species, quasi-steady prediction errors increase as the load ramp rate increases. This is anticipated since highly abrupt ramps result in equivalence ratio deviations from stoichiometric, which undoubtedly influences emissions. Furthermore, it makes sense that load ramp effects would become more substantial as the ramp rate increases. However, for all of the fuel component species, the quasi-steady model performs the poorest for the 2.5 s load ramps and the best for the 1 s ramps. It is difficult to speculate as to why the 2.5 s ramps yield the greatest disparity between experimental emissions and quasi-steady predictions for fuel components. Nevertheless, these results indicate that there are indeed transient, deterministic effects on VOC emissions during load changes, suggesting that the emissions of these species can be predicted with a sufficiently detailed model that accounts for speed/load ramp rate, historical, and equivalence ratio effects. It should be noted that some of the quasi-steady prediction errors may be attributed to additional stochasticity introduced by load transients (i.e., stochasticity statistics may be influenced by transient/historical operation in a similar manner as deterministic emission components). However, this additional stochasticity is likely inconsequential toward the overall conclusion that load transients result in deterministic emissions changes, as most of the statistically significant emission deviations from the quasi-steady model are repeatable, as illustrated in Section A.3 of the Supporting Information document. 3.3. 1,3-Butadiene, Benzene, and Toluene Emissions. The emissions of 1,3-butadiene, benzene, and toluene are notably absent from the analysis above despite the fact that each of these species is identified as toxins by the California Air Resources Board.37 While these species are indeed measured in this work, 1,3-butadiene emissions are completely dominated by stochasticity and minimally influenced by engine load, whereas measurement noise and low compositions prevent a legitimate analysis of benzene and toluene emissions. The chaotic measurements obtained for benzene and toluene are corroborated by the Monte Carlo analysis for estimating measurement noise presented in Section 2.2.1. Estimated measurement noise variances for benzene and toluene are 78.5 and 63.6 ppm, respectively. For reference, the next highest measurement noise variance is 10.6 ppm for pentane, whereas the remaining species lie between 0.4 and 6.5 ppm (Table S1 in the Supporting Information). The exacerbated noise of these species is caused by a combination of low absorptivity, low concentration, and the strong presence of highly

ramps, these trends make sense. Recall that the fuel component emissions during the single positive load ramps also contain a premature peak followed by a temporary downward inflection, whereas the negative load ramp results in an early trough that occurs just midway through the load ramp. Overall, emissions of intermediate species appear to display greater stochasticity than the fuel component emissions for the periodic load profiles, although general relationships with engine operational conditions can still be perceived. Acetylene emissions behave mostly as expected given the single load ramp cases, as emission peaks correspond to load troughs and vice versa. As in the single load ramp experiments, ethylene emissions display interesting patterns that contrast those of other intermediate species. The direct current component (or average) of the ethylene emissions during the periodic load profiles has an inverse relationship with the load ramp rate. Average compositions for the 1 and 5 s load ramps are ∼75 and ∼85 ppm, respectively. This relationship is likely due to the lagged response of ethylene emissions to negative load ramps, as illustrated earlier in Figure 5. Ethylene emissions drop due to the initial positive load ramp and remain at/near the trough for the remainder of the experiment, as emissions fail to rise during subsequent negative load ramps. Interestingly, this effect is more pronounced for the shorter load ramps, revealing a historical effect that is not observed during single load ramps. It should also be noted that ethane emissions, while present in smaller quantities (∼10 ppm), also display similar trends. Similar to the single load ramp cases, formaldehyde emissions display a complicated, albeit repeatable relationship with engine operational conditions during periodic load ramps. The emission profile for the 2.5 s case consists of two moderate spikes surrounding a central, significant upward spike of ∼10 ppm. While the emission profile for the 1 s load ramp case is a simple, stationary wave pattern, the amplitudes of these waves are 5−10 ppm lower than expected given the peaks/troughs of the single load ramp emissions. The emissions during the 5 s ramp case possess the greatest qualitative agreement with the single load ramp data, as peaks occur midway through the positive load transient, and emissions are relatively constant during the negative load transient. While deterministic patterns are present in formaldehyde emissions, clearly a complex model is required to predict them. Methane emissions during the load waves are mostly dominated by stochastic/unrepeatable fluctuations, as expected given previous data. One notable disparity exists between the 1 s periodic and single load ramp emissions. Interestingly, the methane emission peaks for the former (∼90 ppm) greatly exceed those of the latter (∼65 ppm). These disparities may be attributed to equivalence ratio excursions from stoichiometric, which are greater during the periodic load ramps than single load ramps, as will be shown in Figure 10. Thus, these results suggest that methane emissions may be especially sensitive to the equivalence ratio. The standard error between the experimental emissions and quasi-steady predictions for each species during the periodic load ramps is plotted in Figure 8. To determine if the quasisteady prediction errors are statistically significant, the estimated standard stochasticity and standard uncertainty of the experimental emissions are also plotted for each species and load profile. As seen from the figure, the standard prediction errors exceed standard uncertainty and stochasticity in all cases, suggesting that transient and historical effects 5628

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels

3.4. Air-to-Fuel Equivalence Ratio. To elucidate how load transients influence the air-to-fuel equivalence ratio (λ) which impacts emissionsλ is measured for each transient load profile. These measurements are presented in Figure 10.

interfering major combustion products. A detailed discussion of how these effects prohibit an informative analysis of benzene and toluene emissions is provided in Section A.4 of the Supporting Information document. Shown in Figure 9 are 1,3-butadiene emissions for the positive, negative, and periodic load ramp profiles, along with

Figure 10. Air-to-fuel equivalence ratio (λ) measurements during each negative, positive, and periodic load ramp profile.

Figure 9. Emissions of 1,3-butadiene for the negative, positive, and periodic load ramp, 1500 rpm cases. Error bars reflect the standard experimental emission uncertainty, whereas the shaded area surrounding the quasi-steady predictions represents the estimated standard emission stochasticity for the current engine speed/load.

Overall, λ is maintained much closer to stoichiometric during 2.5 and 5 s load ramps compared with 1 s load ramps, although λ has a unique trend for each load profile (negative, positive, and periodic). Pretransient equivalence ratios oscillate between slightly lean/rich values with a period of approximately a second and steady/peak-to-peak sinusoidal components of approximately 0.995/0.03 on average. The equivalence ratio statistics during the 2.5 and 5 s negative load ramps are similar, with steady/sinusoidal components of approximately 1.0/0.03 on average. However, for positive and periodic load ramps, the sinusoidal components for the post-transient λ are slightly elevated with values of 0.045 and 0.035, respectively, which may be partially responsible for the fuel component emissions failing to settle to expected values following positive load ramps. The largest λ excursions from stoichiometric occur during 1 s load ramps. During the negative ramp, λ reaches as high as 1.04 before dramatically falling to 0.95. Positive load ramps result in temporary rich spikes of approximately 0.95. The engine control system encounters even greater difficulty maintaining stoichiometry during periodic ramps, and exceedingly lean λ values of 1.15 are recurrent. These excursions may contribute to the heightened emission deviations from quasisteady predictions during 1 s load ramps (i.e., acetylene, formaldehyde, methane), although emission peaks during single and periodic load ramps are generally in congruence despite differences in λ extremes. One notable exception to this trend is methane emissions, whose peak values extend beyond 90 ppm during periodic ramps, yet reach only ∼60−70 ppm

error bars located near 1 s, which represent the standard uncertainty for the experimental emissions throughout each experiment. Clearly, any deterministic relationships between 1,3-butadiene emissions and engine load are overshadowed by stochasticity and obscured by high measurement noise, as indicated by the sizeable shaded regions and uncertainty bars, respectively. The standard deviation of the stochastic component of 1,3-butadiene emissions is predicted to cover a span of approximately 2 ppm, which mirrors the fluctuations observed in the experimental emissions during the transient load profiles and eclipses the predicted deterministic changes in emissions due to load (∼1 ppm). Furthermore, it is difficult to differentiate between stochastic and deterministic components of emissions due to elevated standard uncertainties, which cover a span of ∼4 ppm. Thus, unlike the species presented previously, it is difficult to form confident conclusions about the 1,3-butadiene emission patterns with respect to load. However, it should also be noted that the composition range covered by the experimental emissions (including uncertainty) spans around 10 ppm, which is modest. Thus, it can be said with a moderate degree of confidence that 1,3-butadiene emissions are overall relatively unaffected by engine conditions and equivalence ratio compared with other species (aside from cyclohexane). 5629

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels

• Steeper load ramps result in greater quasi-steady prediction errors for intermediate species. Interestingly, moderate ramp rates (the 2.5 s ramp cases) result in the greatest deviation for cyclohexane, ethanol, and pentane. • Conclusions cannot be formed for benzene and toluene due to low quantities and excessive measurement noise. • 1,3-butadiene emissions are affected minimally by engine load/speed and appear dominated by stochasticity. • Rich/lean equivalence ratios are more prevalent during 1 s load ramps compared with the remaining load profiles, which may compound quasi-steady prediction errors for these cases.

during single ramps. This suggests that methane emissions may be especially sensitive to equivalence ratio excursions. However, it should be noted that it is difficult to deconvolute λ effects from other transient effects during load ramps. Such a task may require analysis with a neural network (i.e., training two networks on a large collection of data that includes/ ignores λ measurements and comparing the quality of emission estimations between the two networks) and is beyond the scope of this work. Overall, this data suggests that equivalence ratio excursions may contribute toward quasi-steady prediction errors but are unlikely to be solely responsible for them. Significant transient influences are still observed for 2.5 and 5 s load ramps, where λ is maintained close to stoichiometric.

Since VOC emissions are shown to be influenced by transient effects in a deterministic fashion, future work should entail developing a model to predict these emissions, which could play a key role in answering the question of whether control methods can be implemented to mitigate selective VOCs. However, this works shows that relationships between engine parameters and emissions are complicated; the elucidation of these relationships may require a machine learning method (such as a neural network) or a detailed consideration of engine physics.

4. CONCLUSIONS Emissions of various fuel component and reaction intermediate VOCs from a 1.5 L Kubota WG1605 four-cylinder, port-fuelinjected, spark-ignited engine are measured for a collection of transient load profiles. These load profiles consist of single positive, negative, and periodic load ramps spanning durations of 1, 2.5, and 5 s. Emissions are measured using FTIR spectroscopy, and measurements are processed with a previously developed UKF to mitigate the biasing effects of sample recirculation and signal nonstationarity. These emission estimations are compared to predictions from a quasi-steady modelwhich predicts emissions according to current engine speed/loadand estimated magnitudes of stochastic components of emission fluctuations. The data shows that emissions during speed/load transients generally deviate from quasisteady predictions to a degree that cannot be explained by stochasticity alone, indicating that transient effects such as speed/load history and lean/rich equivalence ratios significantly influence VOC emissions. Conclusions from this study are summarized in the following bullet point list. • Greater deviations between the quasi-steady predictions and experimental emissions occur during negative load ramps compared to positive ramps for all species, with deviations greater than 2, 10, and 30 ppm encountered during the former for cyclohexane, ethanol, and pentane, respectively. The quasi-steady model predicts upward fluctuations for these species during the negative load ramps, whereas the experimental emissions drop significantly. • Emissions of the fuel components cyclohexane, ethanol, and pentane are well correlated with one another, displaying peaks/troughs at similar times. • Intermediate species display greater variations in their responses to load transients. Formaldehyde possesses the most complicated, albeit deterministic relationship with engine conditions during transients, as multiple emission fluctuations emerge that result in quasi-steady prediction errors as high as 15 ppm. Ethylene emissions display unique characteristics that include a lagged response to negative load ramps. This characteristic is shared by ethane emissions, which are unreported. • For the periodic load profiles, standard quasi-steady prediction errors surpass the combined effects of stochasticity and uncertainty by statically significant quantities (p > 95%) in 14 out of 21 cases according to reduced χ2 statistics. This indicates the presence of substantial transient and historical effects on VOC emissions.

■

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.energyfuels.9b00350.

■

Quasi-steady maps (Section A.1); emissions at 2000 rpm (Section A.2); repeatability (Section A.3); benzene and toluene emission measurement noise (Section A.4) (PDF)

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. ORCID

David Wilson: 0000-0002-2682-5292 Casey Allen: 0000-0003-3684-5796 Notes

The authors declare no competing financial interest.

■ ■

ACKNOWLEDGMENTS Financial support from Marquette University OPUS College of Engineering is gratefully acknowledged. ABBREVIATIONS BMEP = brake mean effective pressure ECS = engine control system FTIR = Fourier transform infrared ms = millisecond NOx = nitrous oxides ppm = parts per million (on a molar basis) s = second slpm = standard liters per minute UKF = unscented Kalman filter VOC = volatile organic compounds t = time λ = air-to-fuel equivalence ratio

5630

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631

Article

Energy & Fuels

■

air pollutant emissions among mega-cities. Atmos. Environ. 2009, 43, 6435−6441. (18) Brown, S. G.; Frankel, A.; Hafner, H. R. Source apportionment of VOCs in the Los Angeles area using positive matrix factorization. Atmos. Environ. 2007, 41, 227−237. (19) Lau, A. K. H.; Yuan, Z.; Yu, J. Z.; Louie, P. K. K. Source apportionment of ambient volatile organic compounds in Hong Kong. Sci. Total Environ. 2010, 408, 4138−4149. (20) Song, Y.; Shao, M.; Liu, Y.; Lu, S.; Kuster, W.; Goldan, P.; Xie, S. Source apportionment of ambient volatile organic compounds in Beijing. Environ. Sci. Technol. 2007, 41, 4348−4353. (21) Cai, C.; Geng, F.; Tie, X.; Yu, Q.; An, J. Characteristics and source apportionment of VOCs measured in Shanghai, China. Atmos. Environ. 2010, 44, 5005−5014. (22) World Health Organization. WHO Ambient (Outdoor) Air Quality Database: Summary Results, Update 2018; WHO, 2018. (23) Wang, J.; Jin, L.; Gao, J.; Shi, J.; Zhao, Y.; Liu, S.; Jin, T.; Bai, Z.; Wu, C. Y. Investigation of speciated VOC in gasoline vehicular exhaust under ECE and EUDC test cycles. Sci. Total Environ. 2013, 445−446, 110−116. (24) Nakashima, Y.; Kamei, N.; Kobayashi, S.; Kajii, Y. Total OH reactivity and VOC analyses for gasoline vehicular exhaust with a chassis dynamometer. Atmos. Environ. 2010, 44, 468−475. (25) Hong-li, W.; Sheng-ao, J.; Sheng-rong, L.; Qing-yao, H.; Li, L.; Shi-kang, T.; Cheng, H.; Li-ping, Q.; Chang-hong, C. Volatile organic compounds (VOCs) source profiles of on-road vehicle emissions in China. Sci. Total Environ. 2017, 607−608, 253−261. (26) Rakopoulos, C. D.; Dimaratos, A. M.; Giakoumis, E. G.; Rakopoulos, D. C. Evaluation of the effect of engine, load and turbocharger parameters on transient emissions of diesel engine. Energy Convers. Manage. 2009, 50, 2381−2393. (27) Hagena, J. R.; Filipi, Z. S.; Assanis, D. N. In Transient Diesel Emissions: Analysis of Engine Operation During a Tip-in, SAE 2006 World Congress & Exhibition, 2006. (28) Ericsson, E. Variability in urban driving patterns. Transp. Res. Part D, Transp. Environ. 2000, 5, 337−354. (29) Bishop, J. D. K.; Stettler, M. E. J.; Molden, N.; Boies, A. M. Engine maps of fuel use and emissions from transient driving cycles. Appl. Energy 2016, 183, 202−217. (30) Gao, Z.; Conklin, J. C.; Daw, C. S.; Chakravarthy, V. K. proposed methodology for estimating transient engine-out temperature and emissions from steady-state maps. Int. J. Engine Res. 2010, 11, 137−151. (31) Ericson, C.; Westerberg, B.; Egnell, R. In Transient Emission Predictions With Quasi Stationary Models, SAE Powertrain and Fluid Systems Conference and Exhibition, 2005. (32) Pelkmans, L.; Debal, P.; Hood, T.; Hauser, G.; Delgado, M.-R. In Development of a Simulation Tool to Calculate Fuel Consumption and Emissions of Vehicles Operating in Dynamic Conditions, SAE 2004 Spring Fuels and Lubricants Conference, at Toulouse, 2004. (33) Wilson, D.; Allen, C. A Bayesian Estimation Model for Transient Engine Exhaust Characterization Using Fourier Transform Infrared Spectroscopy. Energy Fuels 2017, 31, 11156−11168. (34) Wilson, D.; Allen, C. Experimental Validation of an Unscented Kalman Filter for Estimating Transient Engine Exhaust Composition with Fourier Transform Infrared Spectroscopy. Energy Fuels 2018, 32, 11899−11912. (35) Wilson, D. Bayesian Estimation of Transient Engine Exhaust Composition from Fourier Transform Infrared Spectroscopy Measurements. Dissertation, Marquette University, 2019. (36) Griffiths, P. R.; De Haseth, J. A. Fourier Transform Infrared Spectrometry; John Wiley & Sons Inc.: New York, 1986. (37) California Air Resources Board. Toxic Air Contaminant Identification List [Online] 2011. https://www.arb.ca.gov/toxics/id/ taclist.htm.

REFERENCES

(1) U.S. Environmental Protection Agency. Technical Overview of Volatile Organic Compounds [Online] 2018. https://www.epa.gov/ indoor-air-quality-iaq/technical-overview-volatile-organiccompounds#2 (accessed Aug 15, 2018). (2) Atkinson, R. Atmospheric chemistry of VOCs and NOx. Atmos. Environ. 2000, 34, 2063−2101. (3) Chiemchaisri, W.; Visvanathan, C.; Jy, S. W. Effects of trace volatile organic compounds on methane oxidation. Braz. Arch. Biol. Technol. 2001, 44, 135−140. (4) Carslaw, K. S.; Boucher, O.; Spracklen, D. V.; Mann, G. W.; Rae, J. G. L.; Woodward, S.; Kulmala, M. Atmospheric aerosols in the earth system: a review of interactions and feedbacks. Atmos. Chem. Phys. Discuss. 2009, 9, 11087−11183. (5) Jimenez, J. L.; Canagaratna, M. R.; Donahue, N. M.; Prevot, A. S. H.; Zhang, Q.; Kroll, J. H.; DeCarlo, P. F.; Allan, J. D.; Coe, H.; Ng, N. L.; Aiken, A. C.; Docherty, K. S.; Ulbrich, I. M.; Grieshop, A. P.; Robinson, A. L.; Duplissy, J.; Smith, J. D.; Wilson, K. R.; Lanz, V. A.; Hueglin, C.; Sun, Y. L.; Laaksonen, A.; Raatikainen, T.; Rautiainen, J.; Vaattovaara, P.; Ehn, M.; Kulmala, M.; Tomlinson, J. M.; Collins, D. R.; Cubison, M. J.; Dunlea, E. J.; Huffman, J. A.; Onasch, T. B.; Alfarra, M. R.; Williams, P. I.; Bower, K.; Kondo, Y.; Schneider, J.; Drewnick, F.; Borrmann, S.; Weimer, S.; Demerjian, K.; Salcedo, D.; Cottrell, L.; Griffin, R.; Takami, A.; Miyoshi, T.; Hatakeyama, S.; Shimono, A.; Sun, J. Y.; Zhang, Y. M.; Dzepina, K.; Kimmel, J. R.; Sueper, D.; Jayne, J. T.; Herndon, S. C.; Trimborn, A. M.; Williams, L. R.; Wood, E. C.; Kolb, C. E.; Baltensperger, U.; Worsnop, D. R. Evolution of Organic Aerosols in the Atmosphere: A New Framework Connecting Measurements to Models. Science 2009, 326, 1525−1529. (6) Olumayede, E. G. Atmospheric Volatile Organic Compounds and Ozone Creation Potential in an Urban Center of Southern Nigeria. Int. J. Atmos. Sci. 2014, 2014, 1−7. (7) Atkinson, R. Kinetics and Mechanisms of the Gasphase Reactions of the Hydroxyl Radical with Organic Compounds; American Chemical Society, 1989. (8) Derwent, R. G.; Jenkin, M. E.; Saunders, S. M. Photochemical ozone creation potentials for a large number of reactive hydrocarbons under European conditions. Atmos. Environ. 1996, 30, 181−199. (9) Derwent, R. G.; Jenkin, M. E.; Saunders, S. M.; Pilling, M. J. Photochemical ozone creation potentials for organic compounds in northwest Europe calculated with a master chemical mechanism. Atmos. Environ. 1998, 32, 2429−2441. (10) Butler, T. M.; Lawrence, M. G.; Taraborrelli, D.; Lelieveld, J. Multi-day ozone production potential of volatile organic compounds calculated with a tagging approach. Atmos. Environ. 2011, 45, 4082− 4090. (11) Kampa, M.; Castanas, E. Human health effects of air pollution. Environ. Pollut. 2008, 151, 362−367. (12) Crebelli, R.; Tomei, F.; Zijno, A.; Ghittori, S.; Imbriani, M.; Gamberale, D.; Martini, A.; Carere, A. Exposure to Benzene in Urban Workers: Environmental and Biological Monitoring of Traffic Police in Rome Exposure to benzene in urban workers: environmental and biological monitoring of traffic police in Rome. Occup. Environ. Med. 2001, 58, 165−171. (13) Bahadar, H.; Mostafalou, S.; Abdollahi, M. Current understandings and perspectives on non-cancer health effects of benzene: A global concern. Toxicol. Appl. Pharmacol. 2014, 276, 83−94. (14) Bloemen, L. J.; Youk, A.; Bradley, T. D.; Bodner, K. M.; Marsh, G. Lymphohaematopoietic cancer risk among chemical workers exposed to benzene. Occup. Environ. Med. 2004, 61, 270−274. (15) U.S. Environmental Protection Agency. Health Assessment of 1,3-Butadiene; U.S. EPA: Washington, D.C., 2002. (16) U.S. Environmental Protection Agency. Profile of version 1 of the 2014 National Emissions Inventory [Online] 2017. https://www. epa.gov/sites/production/files/2017-04/documents/2014neiv1_ profile_final_april182017.pdf (accessed: 01 June, 2017). (17) Parrish, D. D.; Kuster, W. C.; Shao, M.; Yokouchi, Y.; Kondo, Y.; Goldan, P. D.; de Gouw, J. A.; Koike, M.; Shirai, T. Comparison of 5631

DOI: 10.1021/acs.energyfuels.9b00350 Energy Fuels 2019, 33, 5620−5631