Impacts and Mitigation of Varying Fuel Composition in a Natural Gas Heavy-Duty Engine.
Low carbon to energy ratio, wide availability and the potential to be generated from renewable sources make natural gas a promising alternative fuel for transportation applications. Its use can be particularly attractive for heavy-duty on-road applications, with high mileage and fixed routes to reduce fueling infrastructure requirements. One challenge for natural gas remains that its composition and performance characteristics are not as tightly controlled as for liquid fuels. Changes in the chemical composition of the natural gas can impact engine performance. This paper assesses the implications of fuel composition variations on a late-cycle direct injection of natural gas engine and presents a combustion sensing strategy that enables a controller to compensate for changes in fuel composition.
Diesel engines dominate heavy-duty on-road applications, due to their high efficiency and high torque. To meet regulatory requirements and customer demands, diesel engine technology has advanced in the direction of improved efficiency and lower emissions. These have been achieved through increases in engine power density, improvements in fuel system architecture, and advances in exhaust aftertreatment. To keep pace with the advances in diesel engines, natural gas technologies are also evolving towards higher power densities. Higher performance natural gas engines are potentially more susceptible to variations in fuel composition. Premixed charge technologies are particularly sensitive to end-gas autoignition (engine knock); many modern premixed NG engines have fuel quality specifications to avoid engine damage due to knock, as well as on-board sensors to detect knock onset and respond appropriately, including through derating the engine if necessary.
One technology that can match diesel engine performance while replacing more than 90% of the diesel fuel with natural gas is high pressure direct injection (HPDI) . HPDI retains a non-premixed, diesel-like combustion process, and hence is not sensitive to knock due to changes in fuel composition. Even so, changes in fuel composition impact the energy content, stoichiometric fuel-air ratio, and density of the fuel, which can impact the performance of an HPDI engine. Changes in concentrations of specific species will also impact engine-out emissions, although the robustness of modern exhaust aftertreatment systems should reduce the impact on tailpipe emissions.
The overall aim of the work reported here was to investigate the impacts of NG composition variation on an HPDI engine. The specific objectives were:
1. Evaluate performance and emissions impacts of natural gas that is representative of the lowest quality that might be delivered to an engine.
2. Demonstrate the ability of a combustion sensing strategy to detect changes in engine performance attributed to natural gas composition variations.
3. Identify whether the combustion sensing strategy could be used as feedback to help to compensate engine performance for very low quality natural gas.
4. Quantify the changes in engine-out and tailpipe emissions after the compensation strategy was implemented.
High Pressure Direct Injection of Natural Gas
HPDI is a pilot-ignited non-premixed natural gas combustion strategy. It is in service in commercial heavy-truck applications in North America and Australia. Results show that maximum torque and efficiency are very close to those of equivalent diesel engines, while more than 90% of the fuel used (on an energy basis) is natural gas . Tank-to-wheel GHG emissions as much as 20% lower than an equivalent diesel engine have been demonstrated, when the HPDI system is optimized for efficiency .
As in a conventional diesel engine, in HPDI the inducted charge consists of fresh air externally mixed with some recirculated exhaust. The charge is then compressed; just before the end of the compression stroke, a small amount of diesel is injected. This is followed by an injection of natural gas through separate nozzle holes. The diesel pilot auto-ignites, providing multiple ignition sources that ignite the natural gas jets. Assessment of key aspects of the HPDI combustion process has been reported previously (e.g. [3,4]).
In an HPDI engine, end-gas autoignition (knock) is not a concern as the fuel and air are not premixed; this allows the engine to retain the diesel engine's high compression ratio. The global air-fuel ratio remains similar to that of an equivalent diesel engine, with similar sensitivity to EGR and combustion phasing. Load control is achieved by simply reducing the amount of natural gas injected; part-load throttling is not required. Emissions of unburned methane are controlled, even at low load, by adjusting the injection process to avoid over-mixing of the natural gas. Crevice and end-gas regions see virtually no fuel, improving fuel conversion efficiency and reducing tailpipe methane emissions. As the combustion is non-premixed, some particulate is formed from the natural gas; this process is particularly sensitive to equivalence ratio .
HPDI FUEL SYSTEM
Liquefied natural gas (LNG) offers benefits of increased volumetric and gravimetric energy density compared to other alternatives for heavy-duty natural gas engines . For HPDI, this state has the added benefit of being substantially less compressible than gaseous natural gas (whether stored at high pressure as compressed natural gas, or in more exotic systems such as absorbed storage), reducing the The diesel pilot is supplied separately, and is compressed using a high-pressure common-rail pump. The diesel is supplied to the engine at a pressure slightly above that of the natural gas. The diesel pressure is regulated using an inlet-metering valve on the diesel pump.
Both high pressure gas and diesel are supplied to the on-engine fuel system. The gas pressure is regulated relative to the diesel pressure using a dome-loaded self-relieving regulator. The two fuels are supplied separately to the injectors; in the first generation HPDI technology reported here, these were delivered through internal rails in the cylinder head. In the latest generation of HPDI technology, the gas and diesel are supplied to the injectors through external fuel rails . HPDI injectors are twin-fuel concentric needle injectors, with separate needles for controlling diesel and natural gas injection through different sets of holes. A schematic of the HPDI injector nozzle is shown in Figure 3. The diesel needle is located inside the gas needle. The diesel holes are located in the tip of the gas needle, which protrudes below the bottom of the injector nozzle. Gas and diesel injection are independently controlled, with separate actuators to enable changes in relative timing and injection durations.
Natural Gas Composition Implications
The implications of variations in natural gas composition on an internal-combustion engine depend on the nature of the combustion event. For premixed-charge spark-ignition engines, higher concentration of species such as ethane and propane in the fuel increases the propensity for end-gas autoignition (knock) . For natural gas engines, this is represented as the 'methane number' (MN) of the fuel, which is roughly analogous to the octane number in gasoline engines . In this scale, pure methane is given a value of 100 and pure hydrogen a value of 0. Any gaseous fuel can be assessed on this scale through testing on a CFR engine running a specific test protocol. To avoid the expense and complexity of testing every fuel blend, calculations are available that provide a reasonable estimate of MN based on measured compositions of gaseous hydrocarbons . Many spark-ignition engines will have a fuel specification, typically around 70-80 MN. Below this level, knock is likely to occur, resulting in significant damage to the engine. Unlike gasoline, the NG available in the commercial distribution network is not typically controlled to a specific methane number; so the risk of providing a sub-optimal or even hazardous fuel to the engine increases.
For non-premixed natural gas combustion, as in the pilot-ignited HPDI technology, there is no premixed charge so end-gas autoignition is not a significant concern. Instead, natural gas composition tends to impact the combustion event through the ignition delay and combustion of the fuel, and the corresponding generation of pollutants. The changes in the energy density of the fuel will also affect the amount of chemical energy available from each injection event. In commercial (non-vehicle) applications, this is evaluated on the basis of the Wobbe Index (WI). The wobbe index is calculated as:
WI = [[HHV]/[[square root of specific gravity]]]
Where HHV is the volumetric higher heating value of the fuel and specific gravity is the density relative to air at standard temperature and pressure (STP). As the fuel pressures in the HPDI system are much higher than STP, the WI does not fully capture the impacts of changes in fuel composition on HPDI engine performance .
A previous study, conducted on a single-cylinder research engine, indicated that increases in either ethane or propane composition could lead to increases in PM and CO emissions . That work was conducted under conditions where all other parameters were artificially controlled, including equivalence ratio. The fuel supply was adjusted to maintain the target IMEP based on in-cylinder pressure transducer feedback. The prior work did not quantify the impacts of changes in fuel composition on brake torque or on the response of the turbocharger system. In particular, the impacts of changing exhaust enthalpy on in-cylinder EQR were not evaluated, although it would be expected to significantly impact emissions.
Ideally, the performance of an HPDI engine would be controlled such that engine hardware was protected and torque was constant independent of the composition of the natural gas fuel.
Measurement of fuel composition would be the most direct way to provide feedback to the engine control (ECU) but at present there are no commercially-viable sensors that would provide reliable readings while meeting the size, durability, and cost requirements for on-road transportation applications.
Indirect assessment of fuel composition can be conducted based on changes in the combustion process, determined from the cylinder pressure trace or from exhaust emissions. Finally, impacts on air-handling system performance can be an indicator of the engine performance. From this, the changes in engine performance caused by variations in fuel composition can be diagnosed and corrected.
In-cylinder pressure transducers can be used to measure cylinder pressure, and from those results the combustion phasing can be calculated. However, the cost and complexity of the sensors and measurement train have limited their application in production environments. An alternative method that has seen significant research is to reconstruct the pressure trace based on indirect measurements. These have included crank kinematics and block or head vibration measurements. Crankshaft speed measurements have been evaluated by a number of researchers (e.g. [11,12,13]); peak cylinder pressure accuracy within 5% has been demonstrated through the use of neural network-based control strategies. Engine vibration measurements, using accelerometers attached to the engine block or head, have also been shown to be able to predict peak pressure magnitude and phasing (e.g. [14,15]). However, this work has not demonstrated the ability to extract more complex information, such as the cumulative heat release.
The work reported here uses a vibration measurement technique, but with the accelerometers mounted on the bearing caps of the engine. This location maximizes the strength of the pressure signal, as the displacement of the bearing cap responds directly to the force exerted by the piston on the crankshaft. It also minimizes engine-to-engine variability, as tighter manufacturing controls mean that vibration transmission through the piston - crank assembly is less susceptible to part-to-part variations than through the cast block or cylinder head. The accelerometer signal can then be combined with crankshaft position sensing to reconstruct the cylinder pressure trace, which can then be used to estimate the apparent heat-release rate (HRR). The knock-sensor based combustion control strategy has been under development for some time [16,17,18,19].
It should be noted that, while the combustion sensing concept is demonstrated here on an HPDI engine with the objective of accommodating variations in fuel composition, it can be equally applied to other combustion strategies. In particular, it could enable advanced combustion strategies where near real-time feedback on cylinder pressure would be beneficial. Part of the work that led to the development of the combustion sensor was as a tool to adjust the diesel injection quantity to control the combustion phasing in a mixed-mode combustion strategy using partially-premixed autoignition . As it is measuring combustion progression directly, it is not sensitive to the fuel or even combustion type.
The work conducted here focuses on a Westport 15L HD multi-cylinder heavy duty natural gas engine equipped with a first-generation HPDI fueling system. This engine was produced commercially between 2007 and 2013, and is certified to US EPA/CARB 2010 emissions standards .
All the tests were conducted on an instrumented research version of the 15L engine. The base engine configuration and calibration were taken from the 2010 certified, 355 kW (475 hp) production engine. Some of the work reported here was conducted on a prototype, higher-power (390 kW) version of the same engine. The research engine was installed on an engine test bed equipped with an eddy-current dynamometer. The specifics of the research engine are given in Table 1. For the 355 kW engine, the base engine hardware (including pistons and valve timing), air handling system, and exhaust aftertreatment are unmodified from the 2007 Cummins 15L ISX engine that the HD 15L was based upon. For the 390 kW version, the pistons were modified to reduce the compression ratio and the injector flow-rating was increased. No other changes were made to the engine. More details on the research engine are presented elsewhere (see [22,23] for the base engine and  for the high power prototype).
The research engine was fully instrumented, including six flush-mounted water-cooled piezoelectric pressure transducers. This data was collected at a 0.5[degrees]CA resolution over 100 consecutive cycles. The in-cylinder pressure from the water-cooled transducers was used to determine the apparent heat release rate (HRR) and to define the combustion phasing. This is defined as the timing at which half of the apparent heat release has occurred, as calculated from the cumulative heat release (CHR). Engine brake torque and speed were determined from sensors on the dynamometer. Intake and exhaust stream pressures and temperatures were recorded at multiple points in the system. Charge air flow was measured using a Meriam laminar flow element and the diesel flow was measured using a gravimetric fuel balance. Emissions of CO, C[O.sub.2], NOx, THC and C[H.sub.4] were measured using a Horiba MEXA3100DEGR emissions bench; this was also used to measure intake C[O.sub.2], to quantify the EGR rate. Engine-out smoke emissions were measured using an AVL 415 smoke meter. These are reported on the basis of filter smoke number (FSN), and only represent black-carbon emissions. Previous work  has shown that percentage reductions in FSN agree well with the reductions determined from gravimetric filter-based measurements.
Repeatability and reproducibility were assessed by evaluating the long-term variations in measured parameters at a fixed operating condition. For this work, the variation in key parameters was evaluated for a reference condition of 2200 N.m and 1490 RPM. The results are shown in Table 2, in the form of coefficient of variation (COV), defined as the standard deviation divided by the mean. The results shown are for variability over a period of approximately three months during the testing program reported here. This value incorporates measurement errors, random variations in engine condition set-point, and long-term variations in engine performance.
Ultra-low sulphur road-grade diesel that met North American fuel specifications was used as the pilot fuel. Natural gas was supplied from a custom fuel supply that allowed various fuel blends to be prepared in advance and then fed to the test cell at the pressures needed for the HPDI engine.
Three fuels were used in the work reported here. The base fuel was commercial gas from the distribution network in Vancouver, British Columbia, Canada ("Line Gas"). This fuel is typical of LNG compositions that are expected to be seen in the field and hence is used as the basis for engine calibration development. The two other fuels were selected to be low-quality gas blends with very high concentrations of heavy hydrocarbons: a high propane blend ("MN65"), and a blend with high levels of both ethane and propane ("MN54"). The MN54 in particular was considered to be an extreme case to test the combustion process and compensation strategies. The compositions of the three gas blends are provided in Table 3. The reference baseline tests were run over all modes with 'line gas' (MN90). The results from this baseline were then used as the reference to evaluate the impacts of the different fuel compositions on engine performance and emissions.
The first fuel (MN65) was selected as being representative of the 'worst case' for natural gas that might be provided through a typical distribution network. Most fueling stations provide higher quality fuels, but it is important to assess the ability of the HPDI engine to operate with such a fuel. The second fuel (MN54) was a particularly ethane- and propane-rich fuel that was selected as being representative of an LNG supply that had weathered severely. It should be noted that typical CNG fuel compositions will be higher than MN75, to avoid knock in modern spark ignited NG engines, while typical vehicular LNG supplies tend to have MN's greater than 85. As a result, fuel quality variations as significant as those evaluated here would be expected to be encountered very rarely in practice.
The testing reported here was carried out over a series of steady-state points representative of the North American Supplement Emissions Test (SET) cycle and European Stationary Cycle (ESC). The specific operating points are provided in Table 4. The operating modes were defined using the torque curve for the research engine, based on the 355 kW Westport 15L HD engine. The engine calibration was taken from the production engine, including setting of fuel pressure, injection timing, and air handling (EGR and VGT position). Over the range of conditions tested, EGR varied from 10-20% by mass, diesel pilot energy ratio varied from about 3% at high load to ~10% at the lowest loads, and combustion phasing (as measured by the mid-point of the cumulative heat release, the CHR50) varied from about 5[degrees] to 18[degrees]CA after top-dead-centre.
Cylinder Pressure Signal Reconstruction
One of the key features of the combustion sensing system is the reconstruction of the cylinder pressure trace from the accelerometer signals. This method was initially developed as a technique to determine the start-of-combustion timing, which was then used as an ECU feedback to adjust the pilot injection event to maintain a desired combustion timing for a partially-premixed auto-ignition combustion strategy . Further development work showed that not only could the start-of-combustion timing be defined, but in fact the pressure trace could be reconstructed during the combustion event. Previous testing demonstrated that the reconstructed pressure trace was reliable up to approximately 20[degrees]CA ATDC. After this point, the vertical component of the force on the bearing cap diminishes very quickly, reducing the signal strength.
For the concept demonstration presented here, the cylinder pressure reconstruction algorithm was implemented in a stand-alone PC using custom code written in C++. The PC used a National Instrument[R] USB 6259 data acquisition system to collect high-speed data from the accelerometers and other engine sensors. The sampling for the accelerometer signal was synchronized with that of the engine crank angle encoder, which has a resolution of 0.5 crank angle degree.
The details of the reconstruction algorithm used in this work were given in previous work . In summary, the effect of a change in pressure in the combustion chamber is transferred through a mechanical transfer path composed of the piston, wrist pin, connecting rod, crank shaft and main crank shaft bearing cap. The mechanical force results in a small deformation of the main bearing cap. The motion of the bearing cap is detected by an accelerometer, which is mounted on the lower surface of the b0earing cap. An equivalent circuit for the accelerometer  is given in Figure 3.
The capacitance Ce and inductance Lm are determined by the mechanical properties of the accelerometer including it elasticity and mass. [C.sub.0] is the static capacitance of the transducer. Ri is the combined leakage resistance of the transducer and resistance of the measurement circuit. The reconstruction algorithm involves solving the governing equations for the mechanical transfer path and the equivalent electric circuit. One of the main challenges for formulating the reconstruction algorithm comes from the fact that the values for most of the above components in the equivalent circuit are not explicitly known or can vary significantly from sensor to sensor. This is overcome by dynamically calibrating the sensor using the known force from the compression signal prior to ignition and a novel method for correcting for charge decay due to leakage resistance [9,10]. Significant effort has gone into simplifying the reconstruction calculations to the point that the process can be completed on a simple chip that can be incorporated into a small stand-alone controller that can communicate over the CAN bus with the ECU. Alternatively, the algorithm could be incorporated directly into the ECU. However, this integration work would be product-specific and was beyond the scope of the reported concept demonstration project.
The fundamental output from the algorithm is a reconstruction of the cylinder pressure in each cylinder on a cycle-by-cycle basis. These can be averaged over cylinders and over cycles to provide a mean cylinder pressure trace. In Figure 4, the reconstructed cylinder pressures at two conditions (mid- and high-load), are compared to the cylinder pressures measured using in-cylinder pressure transducers. The reconstructed pressure traces are the average of 10 consecutive cycles taken from cylinder 2, while the measured pressure traces are the average of 100 consecutive cycles. The figure shows a generally good correlation between the measured and reconstructed pressure traces, although some variations are apparent, especially in the later parts of the combustion and during the expansion process at the higher load case.
The reconstructed cylinder pressure can be used to generate an apparent heat release rate, equivalent to that generated with direct in-cylinder pressure measurement. The HRR from the reconstructed cylinder pressure can be integrated to provide a cumulative heat release (CHR) up to a certain crank angle. The CHR from the reconstructed pressure is compared to the values calculated based on measurements from the in-cylinder pressure transducers for two load conditions in Figure 5. The results demonstrate that the reconstruction tends to diverge later in the cycle; although the 20[degrees]CA limit is somewhat arbitrary, it is evident that in general the errors become larger at later timings. It can also be noted that the bearing-cap mounted accelerometer signals don't reliably identify the initial pilot combustion (starting around -20 to -15[degrees]CA), but generally pick up the main gas combustion reasonably accurately.
Complementary Sensors for Combustion Sensing
To provide a more robust and reliable sensing system, the bearing-cap accelerometer signals can be combined with other sensor inputs. Three parameters that are either commonly measured, or which could be measured through the addition of an extra sensor were checked for correlation with the cumulative heat release rate up to 20[degrees]CA ATDC (CHR20), over the 12 non-idle SET modes. These parameters were intake manifold pressure, exhaust oxygen content, and exhaust temperature. These results are shown in Figure 6, which demonstrates that the strongest correlation was between the cumulative heat release and the intake manifold pressure. The exhaust temperature also correlated well, but it was found to have a long lag due to the thermal mass of the system, and hence was considered to be less well suited to application in a transient engine environment. It is interesting to note that the exhaust oxygen content did not correlate well with the CHR20.
As a result of the strong correlation and fast response of the intake manifold pressure sensor, it was selected to be combined with the accelerometer-based combustion sensor to form the basis of the engine fueling compensation control strategy. Due to the fact that the reconstructed heat release from the bearing cap mounted accelerometer is limited to approximately 20[degrees]CA ATDC, for certain operating conditions, it is not sufficient to obtain the total heat release value. While the intake manifold pressure correlates well with the overall change in the exhaust energy, the latter can be caused by changes in fueling or combustion efficiency in individual cylinders rather than change in the heating value of the fuel. The combustion related issues from individual cylinders, however, can be well diagnosed by the partial heat release rate profile from the reconstructed cylinder pressure. By combining the intake manifold pressure with the pressure-trace reconstruction of the bearing-cap mounted accelerometers, a more robust and reliable detection system can be implemented. This is particularly valuable in conditions with varying ambient conditions or to account for longer-term variation in engine performance, which can be better interpreted from the combined CHR20 and IMP results.
Furthermore, changes in the energy release in individual cylinders could be detected by the CHR20 for that cylinder, but would not be apparent from the IMP measurement. As a result, it is easier to differentiate engine performance variation caused by variations in the composition of the fuel from those caused by changes in either fuel-system or air system performance. Suitable mitigation strategies can then be adopted accordingly.
Sensor Architecture and Implementation
A simplified schematic of the algorithm used by the control strategy is shown in Figure 7. The key measurement inputs to the system are the accelerometer signals from the bearing-cap mounted accelerometers and the intake manifold pressure (IMP). The reconstruction algorithm discussed above generates values for the start-of-combustion (SOC) and the cumulative heat release at 20[degrees]CA ATDC (CHR20). The commanded engine throttle position and engine speed are used to define the expected values of the key parameters, as well as being an important input in the reconstruction algorithm. The structure shown in the figure demonstrates how the recorded information is used to adjust both the fueling timing (start of inject - SOI - correction) and the quantity (gas pulse width - GPW - correction).
To adapt the engine for changes in fuel composition or other changes in fueling behavior, a detailed baseline was first run over the 13 SET modes using standard fuel. The data collected was then used to generate start-of-combustion, CHR20, and IMP maps for the engine. These were then used in the control strategy (outlined in Figure 7) to determine whether observed results at a specific mode with different fuels were deviating substantially from the baseline. The tolerance level for the start of combustion timing (based on 15% total heat release valve) was set at 1 degree; the tolerance levels for the IMP and CHR20 were 5% and 10% respectively. These values were selected conservatively, based on the accuracy and repeatability of the sensors. The concept is to bias towards not making a change in fueling unless a significant deviation is detected. It should be noted that further development work would be expected to reduce these tolerances, and hence improve the precision of the correction. The importance of the IMP as a cross-check to the CHR20 is also shown in the figure; to initiate a change in fueling, both parameters have to change by an amount greater than the defined tolerances.
The corrections are applied to timing or injection quantity separately. If the SOC differs significantly from the expected value, the SOI timing of both diesel and gas injections are shifted together until the SOC falls within the expected range. For the fuel quantity control, if both IMP and CHR20 are outside the desired range, the GPW is adjusted to change the mass of gaseous fuel injected per cycle.
Using the IMP does impose a further challenge on the strategy, in that IMP is a function of other parameters beyond fuel energy added. These factors include ambient temperature and pressure (i.e. altitude) as well as EGR level and residual fraction. As a result, further development for production applications would need to include methods for determining whether a specific IMP was valid beyond a simple map look-up. However, for the purposes of demonstrating the control strategy in a controlled laboratory environment, the look-up table was sufficient.
The impacts of fuel composition variation on the 15L HD research engine were evaluated in three steps. First, for the initial evaluation on the base (355 kW) engine, selected modes were evaluated using the base (MN90) and low quality (MN54) fuels. Then, to provide a cycle-composite evaluation, all 13 modes of the SET cycle were run with the base and intermediate (MN65) fuels. The MN65 fuel was used to avoid exceeding any engine hardware limits when run in 'uncorrected' fueling. The cases were run using the base (production) engine calibration, with a target engine speed and a set 'pedal position' (torque request from the test cell controller). This pedal position defined the injection timing and duration, but did not adapt the fueling commands for changes in actual output torque. The combustion sensing strategy was then implemented and the MN65 fuel SET cycle was repeated with the injection timing and gas pulse width automatically corrected based on the strategy. The bulk of the combustion and emissions evaluation are based on this testing.
To evaluate the impacts of low-quality fuel on a high-power version of the 15L HD HPDI engine, the engine was then reconfigured to achieve 390 kW rated power. The combustion strategy expected values (SOC, CHR20, and IMP) were recalibrated for the higher powers based on stock (MN90) fuel. The engine was then run over the SET modes using the MN90 and MN54 fuels, using the combustion sensing and correction strategy for this higher-power configuration.
Impact of Fuel Composition on In-Cylinder Pressure and HRR
To evaluate the impacts of fuel composition on the cylinder pressure and heat release rate, the research engine was run with fixed throttle position to ensure that the gas pulse width and timing commands were the same independent of fuel composition. The self-trimming features of the intake manifold pressure control were also disabled, so that the VGT position and EGR valve position were not adjusted. This was done to minimize variability in the results.
The outcome of changing the fuel composition from the MN90 used in the baseline tests to very low quality MN54 fuels on the in-cylinder pressure and HRR (as measured by the flush-mounted in-cylinder pressure transducers) is shown in Figure 8 for two operating modes. In interpreting the HRR plots, the initial rise in HRR around 10[degrees]CA BTDC indicates ignition of the diesel pilot. The subsequent much larger increase in HRR shortly after TDC is the combustion of the non-premixed natural gas jets. The impact of the shift in fuel composition is higher torque and later combustion phasing, with indicated power increasing by 15-25% and the combustion phasing (CA50) delayed by 1.5-2[degrees]CA. The general shape of the HRR is not changed, although the late-combustion phase (duration after peak HRR) is increased. Critical parameters extracted from the cylinder pressure data for these conditions are shown in Table 5.
The results shown in Table 5 and Figure 8 demonstrate the impact of changing the fuel composition on the combustion process of an HPDI engine. The higher density of the fuel (as shown in Table 3), when combined with a constant commanded injection duration, means that more fuel mass and hence more chemical energy is injected in each cycle. This is only partially offset by the lower mass-specific heating value of the fuel itself.
Due to the higher enthalpy exhaust, the turbocharger provides more boost, producing higher intake manifold pressures. This results in higher cylinder pressures during compression and combustion, and also leads to higher peak cylinder pressures. However, it is apparent from the pressure trace that the increase in cylinder pressure due to the combustion event (i.e. from the motoring peak to the peak during combustion) is not significantly different for the two fuel compositions. This is a result of the lower quality fuel leading to longer combustion duration rather than a higher peak heat release rate.
The impacts on the combustion process of the MN54 fuel can be observed from the HRR shown in Figure 8. As would be expected, in both cases the pilot ignition has not been influenced by the composition of the gas. The gas start-of-combustion has only been slightly affected, with no discernable influence at A75 and only a small advance at C50. This is a result of the strength of the pilot flame as an ignition source, meaning that the timing of when the gas jets reach the pilot flames dictates ignition timing, rather than the details of the chemical kinetics which would be expected to be influenced by gaseous fuel composition. The combustion duration has been increased in both cases, while the peak HRR has not increased. As the injection duration was fixed, the longer combustion duration suggests that the MN54 fuel is taking longer to mix to a combustible stoichiometry than for the MN90 fuel, possibly due to greater penetration and earlier impingement of the higher density MN54 jet on the piston bowl walls. Detailed optical diagnostics or computational modelling would be needed to clarify this effect, both of which were outside the scope of this study.
Impact of Fuel Composition on Engine Performance
Following the preliminary assessment of fuel composition implications at MN54, the combustion-sensing strategy was implemented in the research engine control system and data was collected over all 13 modes of the SET. In this case, the testing was conducted both with and without fueling correction implemented through the combustion control strategy. To avoid exceeding engine hardware limits, especially at the torque curve, the tests were run on MN65 fuel.
The impacts of the changes in combustion process on torque are shown, over the 12 non-idle SET/ESC modes, in Figure 9. The figure shows that the change in fuel composition increased engine torque, as would be anticipated from the higher density of the fuel. By enabling the sensor-based control strategy to correct the fueling, the torque is returned to closely match the values from the baseline fuel. Deviations in the final torque output are most likely a result of variability in the IMP and combustion sensor measurements: once either IMP or CHR20 were below their uncertainty thresholds, no further correction in fueling was applied. At this stage of development for the system, the tolerance levels for both IMP and CHR20 are relatively broad, leading to the observed variability in torque. Further development effort and improvements in sensor calibration and precision would likely enable further improvement in the accuracy of the fueling correction for torque control.
The increase in fuel mass with the MN65 fuel results in general in an increase in the fuel-oxygen equivalence ratio (EQR), shown in Figure 10. This equivalence ratio is calculated to include oxygen contained in both the fresh air and in the recirculated exhaust gases. The higher EQR is a function of the increased mass of fuel injected for a given pulse-width, due to its higher density, offset by changes in charge mass due to changes in turbocharger response. At all modes, the intake air mass increases significantly with the MN65 fuel, due to the higher exhaust enthalpy and corresponding increase in boost pressure. As with the torque, the correction algorithm brings EQR back to close to its baseline level, as the fuel mass injected is reduced and the lower enthalpy exhaust results in a reduction in turbocharger speed, and hence air flow rate into the cylinder.
Impacts of Fuel Composition on Engine-Out Emissions
The changes in fuel composition can impact emissions results both through the changes to the HRR, to the global equivalence ratio, and through changes to the pollutant formation mechanisms. Previous work highlighted the impact of increased concentrations of ethane and propane on particulate matter emissions at a selected mode in a single-cylinder HPDI engine . The impacts over the full engine map for a multi-cylinder engine have not been reported previously.
The higher torque shown in Figure 9, and longer peak heat release rate shown in Figure 8 combine to result in an overall reduction in power-specific NOx emissions, as shown in Figure 11. NOx emissions are generally consistent across much of the engine map, primarily achieved by adjusting EGR quantity and combustion phasing. In general, reductions on the order of 5-10% are observed with the MN65 fuel, but there is no clear trend with either speed or load. The correction of fueling quantity and timing from the combustion-sensor based control system returns the NOx emissions to close to their baseline levels at most modes. The small impact on NOx demonstrates that the chemical composition of the natural gas has only a small impact on engine emissions across the engine operating map. This is not unexpected, as NOx emissions are dominated by the thermal (Zeldovich) mechanism, which is strongly dependent on local temperature and mixing but is relatively insensitive to fuel species. As the adiabatic flame temperature for the mixture changes by less than 1% between MN90 and MN65 fuels, it is not surprising that no significant impact is seen in NOx once the fueling quantity is corrected. These findings agree with previous research which focused on a single operating condition . At one mode (C75) the fueling correction results in NOx that is significantly higher than for the base MN90 fuel. This is due to lower EGR for the corrected fueling at this point.
In general, power-specific methane emissions decrease with engine load and are consistent across speed; this is shown in Figure 12. For the work reported here, and as is typical in natural gas, changing the fuel quality involves reducing the relative concentration of C[H.sub.4] and replacing it with increasing concentrations of ethane and propane. Inert species ([N.sub.2] and/or C[O.sub.2]) are also added to natural gas to maintain the target Wobbe index, but the impact of inert species was not evaluated here. Reducing the amount of C[H.sub.4] per unit energy in the fuel should result in a reduction in unburned methane emissions. This was generally seen at all modes, as shown in Figure 12. Correcting the fueling only had a marginal impact, which was due primarily to the reduction in the denominator in the power-specific emissions calculation.
The effects of varying fuel composition on CO and PM (as indicated by FSN) show a stronger response than the other emissions, as demonstrated in Figure 13. Mode-by-mode comparison suggests that for the reference fuel, CO emissions increase slightly with speed but are relatively insensitive to load. With the MN65 fuel, CO emissions show a substantial increase, especially at higher loads and higher speeds. The CO emissions return to near-baseline levels when the fueling quantity and timing is adjusted using the combustion-sensing control strategy. This increase with uncorrected fueling is attributed directly to the higher EQR levels shown in Figure 10. As has been seen in earlier work, PM emissions (measured by FSN) increase with higher speeds and loads for a given fuel. With MN65 fuel, FSN levels increased markedly and are only partially corrected by the corrected fueling.
This result demonstrates the importance of heavier hydrocarbons in the natural gas leading directly to higher PM levels; the fact that the CO and NOx return to close to their original levels suggests that the local mixture fractions and temperatures have not been significantly impacted by the change in fuel composition. This is not unexpected, as ethane and propane have easier reaction pathways to [C.sub.2]-containing particle precursors than does C[H.sub.4], which requires recombination reactions; this agrees with earlier single-cylinder engine findings . It is interesting that these increases were seen across the engine operating map, but in particular at the higher-load operating modes.
Low-Quality Fuel in a Higher-Power Engine
The results shown in the preceding sections focused on the MN65 fuel at the base 15L HPDI engine with a power rating of 355 kW (475 hp). For these cases, the results showed performance and emissions were sensitive to fuel composition, but that even with the poor-quality fuel, the engine did not exceed fundamental limits. Impacts on most emissions and torque could be controlled through the use of the combustion sensing strategy.
A follow-on study was conducted to evaluate the impact of very-low quality fuel (MN54 in Table 3), when combined with a prototype higher-power version of the 15L. This engine had a target to increase torque by 10% across the torque curve relative to the 355 kW engine, with a peak rating of 390 kW (520 hp). Along with higher power, emissions needed to remain within US EPA 2010 emission levels. These were achieved through a combination of modified HPDI fuel injectors and reduced compression ratio pistons, as summarized elsewhere .
The higher torque tended to push the enhanced-performance engine closer to mechanical hardware limits than the base 15L with 355 kW rating. As a result, the impacts of degraded fuel quality were more likely to result in conditions with the potential to damage the engine. For this reason, the combustion-sensing strategy was applied to the modified engine to provide engine protection when the MN54 fuel was supplied. The impacts on mode-by-mode torque over the 12 non-idle modes of the SET cycle are shown in Figure 14. The results show that while the torque corrections were not perfect, the variation was relatively small, especially at the highest loads where the potential for engine damage was greatest. These results agree well with the MN65 test results shown in Figure 9.
Over an SET cycle, the impact on emissions of the very low quality fuel was also computed for both engine-out and tailpipe emissions levels. These are shown in Figure 15; it should be remembered that this fuel is significantly lower-quality than the fuels used in certification testing, and in fact is unlikely to be encountered frequently in service. Despite increases in the tailpipe emissions shown in the figure, over the SET cycle, the composite results for all the emissions species remained comfortably below the regulated limits. Tailpipe PM emissions did not show a significant difference, and the levels in both cases were extremely low.
Interestingly, while engine-out NOx was reduced by about 10%, tailpipe emissions increased by 30%. This was attributed to changes in exhaust temperature and composition at select modes impacting SCR performance. It should be noted that no attempt was made to adjust the urea dosing strategy; further optimization of the strategy would likely be able to reduce tailpipe NOx to a level at least similar to the base engine. Tailpipe CO and non-methane hydrocarbon (nmHC) emissions also increased. CO was several orders of magnitude below, and nmHCs were less than 1/3 of the regulated values. In both cases, measurement variability of the low levels of these species post-aftertreatment likely contributed significantly to the observed differences. The exhaust aftertreatment system functioned as designed, significantly reducing concentrations of all the species.
As was shown for individual modes with the MN65 fuel, the main impact of the lower MN fuel was an increase in engine-out PM. As Figure 15 shows, this increased by a factor of 8 in the MN54 case, even with the fueling correction imposed by the combustion-sensing control strategy. It should be noted that the engine out level is total PM measured using gravimetric filters. This did not correspond to any detectable level of FSN post-DPF (the baseline MN90 also has no discernable FSN post-DPF); nor was there any indication of any increase in frequency of regeneration events of the DPF. However, extended operation with fuels very high in ethane and propane would be expected to increase loading on the DPF. If the engine did not spend sufficient time at high loads where the DPF was regenerating passively, an increase in active regeneration frequency would be required. This would be triggered by higher than expected back-pressure levels in the exhaust system upstream of the DPF. The long-term implications of extended operation with the high PM emission levels from low methane content fuels on DPF durability were not evaluated in this work.
A series of tests were conducted on a 15L HPDI of natural gas engine to evaluate sensitivity to low-quality fuels, and to evaluate the suitability of a combustion-sensing based compensation strategy to mitigate the effects of fuel composition variations on engine performance and emissions. The key findings from this work were:
1. HPDI combustion is primarily sensitive to density of the fuel. This is different from modern premixed NG technologies, where the principle sensitivity is to autoignition propensity. The density impacts HPDI combustion by affecting the amount of fuel energy injected in a given gas pulse-width of the injector.
2. Higher concentrations of ethane and propane in the gaseous fuel lead to longer combustion durations but do not significantly impact the start of combustion or the peak heat release rate. In-cylinder pressures increase due to increases in turbocharger speed and boost pressure.
3. A combustion sensor strategy based on intake manifold pressure and bearing-cap mounted accelerometers has been shown to be able to detect changes in combustion energy release. By commanding changes in the pulse-width of the gaseous fuel injection, the controller can adjust for these combustion changes due to variations in fuel composition. The combustion sensing strategy demonstrated here is not HPDI-specific, and could be used for other combustion strategies or other fuels.
4. The main emissions that are affected by fuel composition, once torque is corrected, are engine-out emissions of PM and C[H.sub.4]. C[H.sub.4] is significantly reduced while PM is increased, both due to the increased concentration of ethane and propane in the gaseous fuel. Tailpipe emissions of PM are not impacted due to the high effectiveness of current DOC and DPF aftertreatment systems.
5. Even with a relatively 'low quality' fuel the testing revealed that engine hardware limitations were not exceeded on a standard 355 kW version of the 15L HD engine. Using the combustion-sensing based control strategy enabled a prototype 390 kW variation of the 15L engine to run over a steady-state SET cycle within hardware limits on a fuel composed of more than 35% (by volume) ethane and propane. Tailpipe emissions of all species remained below legislated limits without any optimization of the exhaust aftertreatment system.
[1.] Ouellette P, Goudie D and McTaggart-Cowan GP. Progress in the development of natural gas high pressure direct injection for Euro VI heavy-duty trucks. Presented at ATZ International Motor Congress 2016. Available from: https://www.springerprofessional.de/en/progress-in-the-development-of-natural-gas-high-pressure-direct-/10000924. 2016.
[2.] Dunn ME, McTaggart-Cowan GP and Saunders J. High Efficiency and Low Emission Natural Gas Engines for Heavy Duty Vehicles. IMechE Internal Combustion Engines Conference, London, UK, November 2013. Paper 040.
[3.] McTaggart-Cowan GP, Rogak SN, Hill PG, Bushe WK and Munshi SR. Effects of operating condition on particulate matter and nitrogen oxides emissions from a heavy-duty direct injection natural gas engine using cooled exhaust gas recirculation. International Journal of Engine Research. 5(6). 2004. 499-511.
[4.] McTaggart-Cowan, GP, Jones HL, Rogak SN, Bushe WK, Hill PG and Munshi SR. The Effects of High-Pressure Injection on a Compression-Ignition, Direct Injection of Natural Gas Engine. ASME J. Eng. Gas Turbines Power. 2006; 129(2):579-588
[5.] Jones, H., McTaggart-Cowan, G., Rogak, S., Bushe, W. et al., "Source Apportionment of Particulate Matter from a Diesel Pilot-Ignited Natural Gas Fuelled Heavy Duty DI Engine," SAE Technical Paper 2005-012149, 2005, doi:10.4271/2005-01-2149.
[6.] Bassi, A., "Liquefied Natural Gas (LNG) as Fuel for Road Heavy Duty Vehicles Technologies and Standardization," SAE Technical Paper 201124-0122, 2011, doi:10.4271/2011-24-0122.
[7.] Khalil EB, Karim GA. A kinetic investigation of the role of changes in the composition of natural gas in engine applications. ASME Journal of Engineering for Gas Turbines and Power. 124. 2002. 404-411.
[8.] Kubesh, J., King, S., and Liss, W., "Effect of Gas Composition on Octane Number of Natural Gas Fuels," SAE Technical Paper 922359, 1992, doi:10.4271/922359.
[9.] McTaggart-Cowan GP. Pollutant formation in a gaseous-fueled, direct injection engine. PhD thesis, The University of British Columbia, 2006.
[10.] McTaggart-Cowan GP, Rogak SN, Munshi SR, Hill PG and Bushe WK. The influence of fuel composition on a heavy-duty, natural gas direct-injection engine. Fuel. 89. 2010.
[11.] Gu F, Jacob PJ and Ball AD. Non-parametric models in the monitoring of engine performance and condition: part 2: non-intrusive estimation of diesel engine cylinder pressure and its use in fault detection. Proceedings of the IMechE Part D: J Automotive Engineering. 213(1). 1999. 73-81.
[12.] Potenza R, Dunne JF, Vulli S, Richardson D and King P. Multicylinder engine pressure reconstruction using NARX neural networks and crank kinematics. International Journal of Engine Research. 8(6). 2007. 499-518.
[13.] Mocanu, F. and Taraza, D., "Estimation of Main Combustion Parameters from the Measured Instantaneous Crankshaft Speed," SAE Technical Paper 2013-01-0326, 2013, doi:10.4271/2013-01-0326.
[14.] Zhao, X., Cheng, Y., and Wang, L., "Pattern Recognition Method Applied to Extract In-Cylinder Pressure Excitation Response from Measured Vibration Signals," SAE Technical Paper 2014-01-2703, 2014, doi:10.4271/2014-01-2703.
[15.] Trimbly S, Dunne JF, Bennet C and Richardson D. Unified approach to engine cylinder pressure reconstruction using time-delay neural networks with crank kinematics or block vibration measurements. International Journal of Engine Research. 2016. DOI: 10.1177/1468087416655013
[16.] Huang J and Mumford D. The Development of a Robust Accelerometer-based Start of Combustion Sensing System. DOE-42967-1, 2009. http://www.osti.gov/scitech/servlets/purl/993479
[17.] Huang J, Ancimer R and Chen D. Method and Apparatus for Determining a Normal Combustion Characteristic for and Internal Combustion Engine from an Accelerometer Signal. US patent 8,078,389 B2
[18.] Huang J, Chen D and Mumford D. Method and Apparatus for Using an Accelerometer Signal to Detecting Misfiring in An Internal Combustion Engine. US patent 8,108,131 B2
[19.] Huang J. Method and Apparatus for Reconstructing In-cylinder Pressure and Correcting for Signal Decay. US patent 8,396,649 B2
[20.] Munshi SR, McTaggart-Cowan GP, Huang J, Hill PG. Development of a partially-premixed combustion strategy for low-emission, direct injection high efficiency natural gas engine. Proceedings of the ASME Internal Combustion Engines Fall Technical Meeting, West Virginia. ICEF2011-60181. 2011.
[21.] CARB Executive Order A343-000. Westport 15L HHDD engine emissions compliance certificate. [Online] 06-07-2010. [Cited: Oct 12, 2014] www.arb.ca.gov/msprog/onroad/cert/mdehdehdv/2010/westport_ hhdd_a3430006_14d9_0d20-0d01.pdf
[22.] McTaggart-Cowan GP, Mann K, Huang J, Wu N and Munshi SR. Particulate Matter Reduction from a Pilot-Ignited, Direct Injection of Natural Gas Engine. Proceedings of the ASME Internal Combustion Engines Fall Technical Meeting, Vancouver, Canada. ICEF2012-92162. 2012.
[23.] McTaggart-Cowan, G., Mann, K., Huang, J., Singh, A. et al., "Direct Injection of Natural Gas at up to 600 Bar in a Pilot-Ignited Heavy-Duty Engine," SAE Int. J. Engines 8(3):981-996, 2015, doi:10.4271/2015-01-0865.
[24.] McTaggart-Cowan, G., Mann, K., Wu, N., and Munshi, S., "An Efficient Direct-Injection of Natural Gas Engine for Heavy Duty Vehicles," SAE Technical Paper 2014-01-1332, 2014, doi:10.4271/2014-01-1332.
[25.] Carazo, AV. Novel Piezoelectric Transducers for High Voltage Measurement. Ph.D Thesis, Universitat Politecnica de Catalunya: 242. 2000
Parts of the testing reported here were supported by the California Energy Commission (CEC) under grant PIR-08-045. Support from the research and testing team at Westport, in particular Ken Mann, Alan Uyeno, and Scott Alexander, is acknowledged.
Gordon McTaggart-Cowan, Jian Huang, and Sandeep Munshi
Westport Fuel Systems
Table 1. Research engine specifications Base Engine Cummins ISX15 Bore / Stroke / Con. Rod L. 137/169/261 mm Swept Vol. / Cyl. 2.49 L Number of cylinders 6 Rated power 355 kW 390 kW CR 17:1 15.3:1 Air handling VGT, cooled high pressure EGR Aftertreatment DOC, DPF, SCR Fuel System Injector Westport J36 HPDI Hole number/ angle Diesel: 7/18[degrees]; Gas 9 / 18[degrees] Fuel rail pressures Gas: up to 300 bar Diesel: Gas + 10 bar Table 2. Coefficient of variation of key parameters for the 15L research engine at 1490 RPM, 2400 N.m with base fuel composition. Parameter Long-term COV BSFC [+ or -]1.2% FSN [+ or -]10% HC [+ or -]11% CO [+ or -]8% NOx [+ or -]5% Peak HRR [+ or -]5% Table 3. Primary species in the tested fuels (mole %), and key properties of the fuel Species line gas MN65 MN54 Methane (C[H.sub.4]) 96.4% 84.2% 63.9% Ethane ([C.sub.2][H.sub.6]) 1.9% 4.5% 25.5% Propane ([C.sub.3][H.sub.8]) 0.4% 10.8% 9.6% Butane ([C.sub.4][H.sub.10]) 0.2% 0.2% 0.2% Pentane ([C.sub.5][H.sub.12]) 0.0% 0.0% 0.0% Hexane ([C.sub.5][H.sub.14]) 0.0% 0.0% 0.0% Heptane ([C.sub.7][H.sub.18]) 0.0% 0.0% 0.0% Octane ([C.sub.8][H.sub.18]) 0.0% 0.0% 0.0% Nitrogen ([N.sub.2]) 0.6% 0.6% 0.6% Carbon Dioxide (C[O.sub.2]) 0.2% 0.2% 0.2% Fuel properties Lower heating value (LHV, 49.1 48.3 47.9 MJ/kg) Methane number (MN) 90 65 54 Wobbe index (WI, MJ/[m.sup.3]) 50.5 54.3 56.8 Density at GRP (kg/[m.sup.3]) 193 244 291 Adiabatic flame T (*) (K) 2328 2337 2351 (*)- calculatedfor stoichiometric mixture at 298 K, 1 atm Table 4. Operating modes for the 355 kW 15L HD engine ESC/SET Label Speed Torque mode (RPM) (N.m) 1 Idle 650 0 2 A100 1220 2300 3 B50 1490 1100 4 B75 1490 1650 5 A50 1220 1150 6 A75 1220 1725 7 A25 1220 575 8 B100 1490 2200 9 B25 1490 550 10 C100 1750 1880 11 C25 1750 470 12 C75 1750 1410 13 C50 1750 940 Table 5. Key combustion parameters for two operating conditions comparing MN90 with MN54 with no correction on fuelling. Mode Fuel GIMEP PCP Peak soc CA50 Comb. (bar) (bar) HRR (oCA (oCA dur. (J/[m.sup.3]) ATDC) ATDC) (oCA) A75 MN90 16.1 127 136 13 17.3 32 MN54 18.0 131 142 13.5 18.7 36 C50 MN90 9.1 80 82 9 16.9 33 MN54 11.8 91 85 9 18.7 38