A Predictive Tool for Emissions from Heavy-Duty Diesel Vehicles

Nov 27, 2002 - Data for particulate matter (PM) were available only as a composite value for a whole vehicle test cycle, but using a best effort appro...
51 downloads 11 Views 176KB Size
Environ. Sci. Technol. 2003, 37, 7-15

A Predictive Tool for Emissions from Heavy-Duty Diesel Vehicles NIGEL N. CLARK AND PRAKASH GAJENDRAN* Department of Mechanical and Aerospace Engineering, West Virginia University, Morgantown, West Virginia 26506 JUSTIN M. KERN† Argonne National Laboratory, Building 362, 9700 South Cass Avenue, Argonne, Illinois 60439

Traditional emissions inventories for trucks and buses have relied on diesel engine emissions certification data, in units of g/bhp-hr, processed to yield a value in g/mile without a detailed accounting of the vehicle activity. Research has revealed a variety of other options for inventory prediction, including the use of emissions factors based upon instantaneous engine power and instantaneous vehicle behavior. The objective of this paper is to provide tabular factors for use with vehicle activity information to describe the instantaneous emissions from each heavyduty vehicle considered. To produce these tables, a large body of data was obtained from the research efforts of the West Virginia University-Transportable Heavy Duty Emissions Testing Laboratories (TransLabs). These data were available as continuous records of vehicle speed (hence also acceleration), vehicle power, and emissions of carbon monoxide (CO), oxides of nitrogen (NOX), and hydrocarbons (HC). Data for particulate matter (PM) were available only as a composite value for a whole vehicle test cycle, but using a best effort approach, the PM was distributed in time in proportion to the CO. Emissions values, in g/sec, were binned according to the speed and acceleration of a vehicle, and it was shown that the emissions could be predicted with reasonable accuracy by applying this table to the original speed and acceleration data. The test cycle used was found to have a significant effect on the emissions value predicted. Tables were created for vehicles grouped by type (large transit buses, small transit buses, and tractor-trailers) and by range of model year. These model year ranges were bounded by U.S. national changes in emissions standards. The result is that a suite of tables is available for application to emissions predictions for trucks and buses with known activity, or as modeled by TRANSIMS, a vehicle activity simulation model from Los Alamos National Laboratories.

Introduction Accurate prediction of emissions inventory is important for appropriate apportionment of species to sources and for assistance in projecting the effect of changes in vehicle technology and in transportation. This study was motivated by the need to augment a traffic simulation model, TRANSIMS * Corresponding author phone: (304)293-3111, ext. 2490; fax: (304)293-2582; e-mail: [email protected]. † Previously M.S. student at West Virginia University. 10.1021/es0113192 CCC: $25.00 Published on Web 11/27/2002

 2003 American Chemical Society

(Los Alamos National Laboratories), with emissions prediction capability, but the approach has wide and general application. It was necessary to produce an emissions prediction tool that accepted speed and acceleration as inputs. Emissions prediction matrices for oxides of nitrogen (NOX), carbon monoxide (CO), carbon dioxide (CO2), and hydrocarbons (HC) were prepared for different vehicle model years. Prediction of particulate matter was achieved by proportioning known PM emissions with respect to speed and acceleration using CO emissions. The database was generated using the West Virginia University-Transportable Heavy Duty Vehicle Emissions Testing Laboratories (TransLabs), which are described below. These two laboratories have collected data across the nation primarily on performance of alternately fueled trucks and buses for the U.S. Department of Energy. In doing so, substantial information was gathered from diesel control vehicles. Vehicle emissions were characterized using a variety of driving cycles, including the CBD Cycle, 5-Peak Cycle, 5-Mile Route, NY Bus Cycle, and the CSHVR (1, 2). The existing data were used in an analysis to develop a method that can produce emissions factors in grams per second for all heavy-duty vehicles from a small database of measured emissions. The method developed categorizes the emissions according to the vehicle speed and acceleration. These two laboratories have collected data across the nation primarily on performance of alternately fueled trucks and buses for the U.S. Department of Energy. In doing so, substantial information was gathered from diesel control vehicles. Vehicle emissions were characterized using a variety of driving cycles, including the CBD Cycle, 5-Peak Cycle, 5-Mile Route, NY Bus Cycle, and the CSHVR (1, 2).

Technical Description of Laboratory The two West Virginia University Transportable Heavy Duty Vehicle Emissions Testing Laboratories (TransLabs) are heavy-duty chassis dynamometer systems that can be moved from site to site with a dedicated semitrailer and a laboratory trailer. The cycle averaged emissions data gathered by the laboratories are added to a database (http://www.afdc. nrel.gov/web_view/emishdv.html) maintained by the National Renewable Energy Laboratory (NREL), in Golden, Colorado. The inertia and road losses, including wind drag and rolling resistance, are simulated using selectable flywheels and air-cooled eddy current power absorbers. Power is taken directly from the drive-wheels of the tested vehicle via hub adapters while the vehicle runs on free-spinning rollers. Hub torque, vehicle speed, engine speed, and gaseous emissions data can be logged continuously during a test through use of a full scale exhaust dilution tunnel, with heated probes and sample lines and analyzers for carbon monoxide (CO), oxides of nitrogen (NOX), and hydrocarbons (HC). Particulate matter (PM) is determined gravimetrically by collecting the PM on 70 mm diameter filters. These two laboratories have previously been correlated with one another, and both laboratories were used to collect data that were used in this analysis.

Methods of Generating Emissions Factors Various approaches exist or can be suggested for the prediction of heavy-duty vehicle emissions contributions to the national or to a regional inventory. In the absence of accurate measurement of emissions from every vehicle performing every task, all approaches must be an approximation, and it may prove impossible with current data VOL. 37, NO. 1, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

7

to develop highly accurate approach. Following are some of the approaches that can be used for emissions prediction. 1. Use of Certification Data 2. Direct Use of Chassis Dynamometer Data 3. Use of Power Based Emissions Factors 4. Use of NOX/CO2 Ratios 5. Use of Modal Approaches 6. Use of Speed-Acceleration Data Use of Certification Data. The present and previous U.S. Environmental Protection Agency (EPA) approaches to prediction of emissions output from heavy duty vehicles have relied on the use of emissions certification data to yield, for each species, an emissions factor with units of grams per brake horse power-hour (g/bhp-hr). This is the approach used in EPA’s MOBILE5 and MOBILE 6. This approach was also used in California Air Resources Board’s (CARB) EMFAC 7. The Federal Testing Procedure (FTP) is a transient stationary dynamometer test used to evaluate an engine’s emission production level for federal certification and is used to provide emissions input for MOBILE. The target values (engine speed and load values) were arrived at through the use of a Monte Carlo simulation of data collected in Los Angeles, CA and New York, NY in the 1970s (3). Emissions factors for use in models such as MOBILE for heavy-duty vehicles are usually expressed in grams per mile (g/mile). The emissions certification data for the engines that power these vehicles are in units of grams per brake horsepower-hour (g/bhp-hr). One method of converting the certification data to vehicle emissions is to use the formula from Machiele (4).

g g lb bhp - hr gallons ) mile bhp - hr gallon lb mile

(

)(

)(

)(

)

(1)

This approach may be criticized because the FTP emissions certification test is based on vehicle behavior that is probably not relevant to today’s vehicle usage and certainly cannot represent the extremes of freeway cruising and stopand-go city service vehicle behavior. As an example, vehicles today cruise with engines at a lower speed and are less “gearbound” than the FTP represents. Certification data may also not reflect emissions in the field if “off-cycle” injection timing strategies are enabled in the engine controller. The off-cycle timing can lead to substantial increases in NOX and minor reductions in PM and fuel consumption and are typically invoked during rural or cruising operation. The off-cycle emissions are clearly evident in previously published NOX vs power plots (5). Although off-cycle emissions will be curtailed in the future, they are present in many diesel vehicles manufactured over a decade of model years. Degradation factors are provided by the manufacturers to describe the change in certification test emissions with respect to accumulated mileage on engines, but very few data exist to show how emissions decline in real use. These factors do not account for engines that have endured tampering or malfunction or are approaching retirement or rebuild age. Some effects of tampering with engine hardware have been described by McKain et al. (1). Direct Use of Chassis Dynamometer Data. Heavy-duty vehicles may be subjected to emissions characterization on a chassis dynamometer, as is the present approach for light duty vehicles (6-8). The emissions results may be obtained in g/mile for each emissions constituent. A simple approach for prediction therefore involves taking the product of these emissions factors and the vehicle miles traveled. This approach is at least as valid as the present MOBILE approach and offers the advantage that fuel economy need not be considered in the process. All else being equal, a vehicle with a less efficient drivetrain would simply yield higher emissions factors in units of grams per mile (g/mile). The 8

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 1, 2003

California Air Resources Board has moved to this method for inventory of heavy-duty truck and bus emissions in EMFAC 2000. There is the advantage that vehicles subjected to chassis dynamometer emissions characterization can be tested as received, including influences of tampering or malfunction that might be lost if the engine were first removed from the vehicle. An example would include the influence of a clogged air filter. There is also the advantage that vehicles are more readily tested using chassis dynamometer systems than by removing the engine from the vehicle, so that data more representative of the whole fleet, rather than new vehicles, can be obtained. A problem arises in that no single cycle can hope to represent the real world spectrum of vehicle activities. Although the Urban Dynamometer Driving Schedule (TestD) (9) exists as a companion to the engine certification test, it does not correlate well with the FTP engine certification test (8) and does not represent all behaviors. The test cycle used has significant effect on the emission levels. A portion of testing at WVU under a recent contract for the National Renewable Energy Laboratory (NREL) involved testing a single truck on many different test cycles. It was found that the emissions levels were influenced by the test cycle used. The NOX data in grams per mile for a variety of test schedules are shown in Figure 1 from Nine et al. (2). Similar variation in emissions results was obtained for the CO and PM emissions. The vehicle was a 1995 GMC box truck with a Caterpillar 3116 engine rated at 170 hp. The fuel used was D2 diesel, and the vehicle had a test weight of 22 000 pounds. It is concluded that the approach of using direct chassisbased emissions factors is acceptable only if the test schedule is sufficiently representative or if correction factors are employed. Power Based Emission Factors. Chassis dynamometer data need not be employed directly as emissions factors. During most chassis dynamometer testing, continuous emissions data are acquired for NOX, CO, and HC, and these can be considered in the development of models that can then be used to project the emissions from the vehicle under a broad range of operating conditions. These data are taken in units of parts per million of species in diluted exhaust but are readily converted to units of grams per second using the dilution tunnel mass flow rates. It is possible, if a successful model can be developed to relate the emissions from a vehicle to its operating parameters, that the emissions may be predicted for any other cycle for which the operating conditions are known. Ramamurthy et al. (5) have espoused an approach where the emissions are related to the instantaneous power output from the vehicle rear axle using the continuous power and emissions data. Instantaneous chassis dynamometer emissions data for a particular vehicle are processed to yield the instantaneous emissions in grams per second as a function of a single variable, rear axle power, as shown in Figure 2. In using these factors, one must employ correct time alignment of instantaneous power and the emissions constituent (10). Work has been completed at WVU to understand the time alignment of instantaneous power and its resulting emission production. The axle power is measured instantaneously, yet the resulting emissions are measured after a time delay of the gas traveling from the engine to the analysis bench. Also, in addition to time delay, an effect defined as “smearing” or time diffusion of the gaseous emissions data occurs. This arises from axial mixing in the exhaust transfer tube, dilution tunnel, and sampling lines and from the response of the analyzers. For example, in the case of infrared analyzers, there is time needed to refill the measurement cell with the sample gas. Clark et al. (11) showed experimentally that the measurement response to a spike output of emissions from an engine

FIGURE 1. NOX emissions of various test schedules on a box truck (1).

FIGURE 2. Continuous NOX versus power for a 1996 transit bus tested on the CBD cycle. resembles a gamma distribution. Typically for a spike input, 90% of the emissions are detected in a 12 s period. Ramamurthy et al. (5) have found the approach of relating emissions to axle power to be successful for modeling of NOX emissions (and for carbon dioxide, which is not regulated) from diesel vehicles but difficulties arise for the cases of CO and HC. Also, the approach cannot account for “off-cycle” injection timing, which is discussed elsewhere in this paper. Although modeling of gaseous emissions is possible because continuous data are available, PM is measured gravimetrically, as a composite for the whole test, so that instantaneous PM is not known. The best available tool at present involves proportioning the lumped PM over the duration of the cycle in linear proportion to the CO. There is sympathy in PM mass production and CO production although lubricant contribution, heavy HC contribution, sulfate contribution, and varying PM formation mechanisms all cause this to be an approximate approach (12). In the future, data from devices such as the Tapered Element Oscillating Microbalance (TEOM) may provide more accurate estimates of instantaneous PM emissions. In addition, Neural Net models are showing promise in the modeling of diesel species in cases where simple models fail (5, 13-15). Neural nets employ weighted combinations of input variables that

are passed through nonlinear activation functions. Neural nets may have many such layers and are therefore capable of complex inferences of emissions based on appropriate inputs. NOX/CO2 Ratios. NOX and CO2 values on a mass basis (grams) can easily be reported as a NOX to CO2 ratio. Fuel usage on a volume basis (gallons) can be inferred from the CO2 mass production thus giving a predictor of NOX emissions produced on a per gallon of fuel basis. The CO2 production, or mass of CO2 produced per mass of fuel used, can be obtained by using a carbon balance. This results in 44 g of CO2 produced for (approximately) every 13.8 g of diesel used. Again, the database values are subject to the characteristics of the driving cycle followed on the chassis dynamometer. Example of NOX/CO2 Emissions Factors. To demonstrate this method, data from a 1996 transit bus powered by a Cummins M-11 engine rated at 280 hp were used. Figure 3 shows the continuous data of NOX plotted against CO2 for 5 consecutive test runs on the CBD Cycle. Data scatter arises due to the “smearing” of instantaneous values by the two analyzers coupled with the presence of severe transients in the cycle. The average value of the ratio throughout this testing was determined to be 0.014 g of NOX per gram of CO2. To predict vehicle emissions in grams per mile, the fuel mileage of the vehicle, density of the fuel, and CO2 production VOL. 37, NO. 1, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

9

FIGURE 3. Variation of NOX with CO2 for 5 consecutive test runs on the CBD cycle for a 1996 model year 40-foot transit bus powered with 280 hp Cummins M11 engine. per amount of diesel need to be determined. Then eq 2 can be used to obtain the emissions value in grams per mile.

gNOx ) mile 0.014 gNOx 44 gCO2 gallon 3217 g diesel (2) gCO2 13.8 g diesel 5 miles gallon

(

)(

)(

)(

)

This approach will not prove sufficiently accurate if the engine timing varies substantially over the operating envelope. It will also be unreliable for other emissions species, which do not vary in proportion to CO2. Modal Approaches. In addition to the continuous data approach, segments of the test may be considered, yielding modal emissions factors. It is argued that any vehicle behavior can be viewed as a collection of modes such as “cruising at high speed”, “idling”, or “accelerating rapidly”. This approach exists as a simplification of the modeling approach, but it is argued that it will at best be approximate when considering response of PM, CO, and HC to transient engine behavior. Use of Speed-Acceleration Data. This approach is closely related to the modeling and modal approaches and is the approach discussed in detail in this paper. It is common in reviewing light-duty vehicle emissions data to consider the speed and acceleration of the vehicle to be governing independent variables. For a given vehicle, the speed governs the road load losses, and the product of speed and acceleration governs the instantaneous inertial power demand. Emissions for a vehicle can be binned according to its speed and acceleration characteristics in the postprocessing of cycle data. There is a question as to whether vehicle speed and acceleration offer advantage over the single variable of power in heavy-duty applications, since the engine responds solely to power demand, and vehicle acceleration rates are low in heavy-duty vehicles. However, since more gear shifting occurs at lower speeds, speed is likely to add value as a variable. The objective of the emissions model is therefore to provide an emissions value, in the units of g/mile or g/sec, for each species as a function of speed and acceleration. This is accomplished by placing measured instantaneous emissions data into predetermined speed and acceleration bins and averaging the data in each bin. Problems in using a speed-acceleration approach for prediction arise when the speed-acceleration profile of the vehicle for which an emissions factor is to be determined 10

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 1, 2003

FIGURE 4. Speed versus acceleration for a 1996 model year transit bus driving the CBD cycle. encounters hills or grades. The extent to which a grade affects the emissions is not well-known because the test schedules used to date on chassis dynamometers have no provisions for simulating hills. The WVU chassis dynamometers do not have the ability to motor the vehicle to simulate downhill driving and are limited in their ability to absorb full power at low speeds. This presents a problem when correlating the emissions to the speed-acceleration profile of the actual activity of a vehicle. As a vehicle is traveling uphill, the rate of change of speed (acceleration) is low, while the axle power demand is high as compared when the vehicle is traveling on level ground, as simulated on the dynamometer. The only full-power emissions data that are gathered on the dynamometer are at a high rate of change of speed. This means that the predicted emissions of the vehicle ascending the grade will be lower than the actual emissions produced. This will hold true for emissions species that correlate well with axle power, such as NOX. Likewise, the emissions predicted when the vehicle descends a hill will be higher than the actual emissions produced because the vehicle can attain a relatively high rate of change of speed, for which there is emissions data at full power. Unfortunately, not all existing cycles cover the speedacceleration envelope thoroughly. Figures 4 and 5 show the speed vs acceleration plots for a 1996 transit bus powered by 280 hp Cummins M-11 engine over a CBD Cycle and a 1998 transit bus powered by a 280 hp, Cummins M-11 engine

TABLE 1. Acceleration Bin Ranges bin label heavy acceleration medium acceleration light acceleration cruise light deceleration medium deceleration heavy deceleration

FIGURE 5. Speed versus acceleration for a 1998 model year transit bus driving the CSHVR. over the City-Suburban Heavy Vehicle Route (CSHVR), respectively. The CBD has all acceleration rates defined, whereas the CSHVR is a speed-distance based route, and at certain points require maximum vehicle acceleration. It can be seen that the CBD Cycle fails miserably in covering the envelope, although the CSHVR has better coverage. The WVU database, which was used in this research, consists of individual test runs of heavy-duty vehicles that were driven on different test sequences. The term “test sequences” encompasses both routes and cycles that are present in the database (11). The test runs contain emissions data on a second by second basis. As the exhaust gas leaves the vehicle, it travels through a transfer pipe, into the dilution tunnel, and then to the analyzers where the emissions are measured. The resulting time shift of the emissions data must be corrected to be able to correlate a particular speed/acceleration of the vehicle to an emissions event. This was accomplished using a cross-correlation method employing eq 3

S)

d(load) d(ES)(t + ∆t) ‚ dt dt



(3)

where “load” (power in HP) and “ES ” (grams/sec) represent the continuous curves of the emissions species recorded from the test. Different parameters were used for the load on the vehicle including axle power, axle torque, and vehicle acceleration. The axle power was chosen for the correlation. The sum “S” is calculated for different values of the time shift, ∆t, and the time shift that produces the largest sum is used as the best correlation. This process is repeated for each of the emissions species to obtain the corresponding value for the time shifts for each of the species. This method has been used previously by Messer and Clark (10), and the authors have found that cross-correlating the first differentials is superior to cross-correlating the signals themselves. The different speeds and accelerations that a heavy duty vehicle experiences during travel must be divided into bins so that any operating point defined by a speed value and an acceleration value is represented by a specific bin. The emissions factors are then placed in the appropriate bins based on the corresponding speed and acceleration from the test data. There may be many operating points in each of these bins. An average value is evaluated and is placed in each bin. The speed data were divided into 0.5-mph bins, and the acceleration data were divided into seven ranges: Table 1 shows the acceleration value ranges chosen for each bin. Verification of Speed-Acceleration Approach. Verification of the speed-acceleration approach for NOX was done in the following manner. The speed-time trace for a cycle (in this case for a CBD cycle) was selected for the analysis. The

acceleration range >2 mph/s 1 to 2 mph/s 0.3 to 1 mph/s -0.3 to 0.3 mph/s -0.3 to -1 mph/s -1 to -2 mph/s 2 mph/s

medium accel. 1 to 2 mph/s

light accel. 0.3 to 1 mph/s

cruise -0.3 to 0.3 mph/s

light decel. -0.3 to -1 mph/s

medium decel. -1 to -2 mph/s

heavy decel. 0.3 mph/s -0.3 to 0.3 mph/s