Home   |    International   |    Contatti   |    Info   |   


Combustion Noise Prediction in a Small Diesel Engine Finalized to the Optimization of the Fuel Injection Strategy

cnr

Fig. 1 - Engine and test-bench laboratory

The worldwide demand for the engine optimization, in terms of power output, produced pollutants and fuel consumption, is continuously increasing [1]. A growing attention is also being devoted by automobile manufacturers to NVH characteristics of the whole vehicle, and, for this reason, the control of the noise produced by the combustion process in a diesel engine is being considered a very important topic [2,3]. For this reason, the combustion noise reduction is nowadays considered as an additional factor in engine development alongside performance, fuel consumption and emissions. The engine under investigation in the present work is a naturally aspirated, light-duty diesel engine, equipped with a mechanical Fuel Injection System (FIS) and utilized in non-road applications. It is currently under development; however, a new prototype equipped with a common-rail (CR) FIS, going to be installed within small city cars. It is well known that the use of a CR-FIS gives the possibility to respond to the noise emission legislation and market demand through modulation of the injection parameters. The above improvements are however more difficult to obtain on small displacement engines, because of the complexity and cost of the FIS itself [4,5]. In addition, a long development phase is usually required at the test bench in order to define the optimal injection strategies in different engine operating conditions.

Based on the above considerations, the main scope of the present work is the development of an optimization procedure that is able to theoretically determine the best injection strategies compatible with high performance and low noise levels, while reducing the development phase and the time-tomarket of the engine. To fulfill the above goal, a multi-objective optimization tool was employed [6-8]. This tool was able to automatically vary the control parameters, and to compare the related performances. The optimization tool however required the development of proper simulation models of the engine. It is currently possible to simulate the physical and chemical processes occurring in the operation of internal combustion engines by using appropriate numerical codes (1D or 3D). 3D simulations can predict, for example, spray behavior, mixture formation, combustion process and toxic emissions [9]. However, they require high computational times, even when high-speed computers are employed. 1D models, on the contrary, are able to gain information on the overall engine behavior [10] and, due to the reduced computational efforts, are better suited to be employed within an optimization procedure [11].

Automotive
Altri case studies sullo stesso argomento

Ottimizzazione
Altri case studies sullo stesso argomento

Calendario 2016
Corsi di formazione sulle tecnologie software

Formazione personalizzata
Percorsi formativi personalizzati e traning-on-the-job: contattaci!

Workshop e seminari
Visita la pagina degli eventi: potrai partecipare alle iniziative EnginSoft ed essere informato sui più importanti appuntamenti europei

Newsletter EnginSoft
Leggi l'ultimo numero ed Abbonati alla rivista!

X

Chiedi all'esperto!
Articolo completo

cnr

Fig. 2 - In-cylinder pressure and accelerometer signal

Concerning the prediction of noise emission, either detailed or simplified models can be utilized. Detailed approaches are usually based on the employment of FEM-BEM codes, which include the in-cylinder pressure cycle as an excitation on the engine structure [12,13]. A number of alternative and simplified procedures are also available in the current literature, based on a proper processing of the computed pressure cycle [14-16]. In this paper, a 1D model was chosen to predict the engine performance and a recent methodology [14], based on the decomposition of the 1D computed pressure signal, was utilized to estimate the combustion radiated noise. The whole activity was developed in four main steps:

  1. an experimental campaign was initially carried out to gain information on performance and noise levels on the engine and to acquire the data required to validate the 1D and the combustion noise models;
  2. the 1D simulation of the tested engine was combined with the GT-Power® code [17], for estimating the in-cylinder pressure cycles and the overall performances;
  3. the methodology reported in [14] was included within a Matlab® routine to estimate the noise level. Some coefficients included in the above correlation were properly tuned to be in agreement with the experimental data;
  4. an optimization process [18] was finally carried out with the modeFRONTIER® code to identify an optimal injection strategy of the prototype engine equipped with the CR-FIS. The objectives established were the maximization of the engine performance and the reduction of the noise emission, at a constant load and rotational speed.

cnr

Fig. 3 - 1D scheme of the tested engine in GT-Power

In the following, a description of each of the above steps is presented and some conclusions are finally drawn concerning, in particular, the trade-off between the two objectives, requiring the selection of a compromise solution. The latter was identified through the employment of a “Multi-Criteria Decision Making” (MCDM) tool, provided by the modeFRONTIER® code.

Experimental Analysis
In this study, a naturally aspirated, four stroke, two valve, single cylinder diesel engine (505 cm3 displacement) was experimentally investigated. The engine test bed included an electrical dynamometer, the data acquisition and control units, as well as emission and an acoustic measurement equipment. PERFORMANCE TESTS - A programmable electronic control unit (PECU), based on a dSpace processor, was used to manage the engine operating conditions. The in-cylinder pressure was detected by a piezoelectric pressure transducer, connected to the AVL IndiModul 621. The air flow rate was also estimated through the measurement of the fuel flow rate and Air/Fuel Ratio. The latter was derived from the analysis of the exhaust gas composition. Engine tests were carried out at full load, in a range of engine speed going from 1400 to 3000 rpm. ACOUSTIC TESTS - In order to measure the radiated noise with reasonable accuracy, the acoustic characteristics of the environment must be known. On an acoustic basis, the ideal environment is a space with no reflecting surfaces and no background noise. In practical terms the ‘best’ environment, for engine applications, is an open-air site with one hard reflecting surface (the ground) and no other obstructions for at least 50m from the noise source and microphone positions. Moreover, the background noise level should be at least 10 dB (preferably 20 dB) below the measured one.

Engine noise measurements could be also made in ‘nonanechoic’ test cells with acoustic absorption. The latter set-up was followed for the present investigation, where certain important parameters (reverberation time of the room and background noise) were taken into account, as prescribed by the applied ISO Standards [19]. For this aim, the engine was acoustically isolated from the electrical dynamometer, through properly designed acoustic absorption panels (see the white shields shown in figure 1).

The radiated noise of the engine surfaces was measured by a free-field microphone located 1 m away from the engine block, to avoid near field effects, and in a position away from the intake and exhaust systems. The purpose was to minimize the effects of flow noise sources, so that the major contribution considered is the one coming from the engine block. The calibration of the microphone was performed before each test by means of a pistonphone. Simultaneously, an accelerometer was positioned on the engine head in order to correlate the vibration, microphone and pressure signals. The acquisition of the noise and vibration signals were performed at all investigated engine speeds, at sampling frequencies of 48 kHz. In this way a useful bandwidth wider than 20kHz, free from aliasing effects, was available. In order to synchronize the measurements, a pulse signal supplied by an optical encoder was used as a reference.

cnr

Fig. 4- Comparison on the pressure cycle at 1400 rpm

cnr

Fig. 5 - Comparison on the pressure cycle at 2200 rpm

cnr

Fig. 6 - Comparison on the pressure cycle at 3000 rpm

cnr

Fig. 7 - Computed and experimental air flow rate

cnr

Fig. 8 - Computed and experimental mechanical power

cnr

Fig. 9 - Total pressure spectrum (in dB), at 1400 rpm

cnr

Fig. 10 - Decomposition of the total pressure in motored, combustion and resonance contributions, at 1400 rpm

cnr

Fig. 11 - Comparisons on the overall noise

cnr

Fig. 12 - Parametric Injection strategy

Figure 2 reports an example of the acquired accelerometer signal, phased with the in-cylinder pressure, during both combustion and motored conditions, at 3000 rpm. In both cases, a strong vibration peak is well evident at the top dead center, due to the piston slap phenomenon. Some reduced spikes can also be identified, in motored conditions, when the incylinder pressure approximately crosses the crankcase pressure (around the atmospheric level). This is probably an indication of the occurrence of some piston movement around its pin, captured as a small vibration signal. The same does not happen when the piston is loaded by combustion pressure. The pressure profiles also show the presence of a strong disturbance at the inlet valve closure (IVC) event. The same spike is indeed absent on the accelerometer signal.

Additional experimental data will be presented in the following sections in comparison with the numerical results.

One-Dimensional Simulation
The well-known GT-Power one-dimensional simulation code was employed to predict the performance of the investigated engine, schematized as shown in figure 3. The 1D code solves the mass, momentum and energy equations in the ducts constituting the intake and exhaust system, while the gas inside the cylinder is treated as a zero-dimensional system. Concerning the modeling of the combustion process, a classical Wiebe equation was utilized to compute the heat release rate in the base engine. Proper values of the combustion process duration during both premixed and diffusive phases were specified. Figures 4-6 show the comparison of the computed and experimental pressure cycles at three different engine speeds. The model was able to correctly reproduce the pressure evolution along both the compression, combustion and expansion phases. Figures 7 and 8 instead report the comparison of the inlet air and brake power through the whole range of investigated engine speeds. Once again a good agreement between experimental and numerical results was reached, especially concerning the air flow rate (fig. 7). The good matching obtained with the experimental data allowed the extension of the 1D analysis to the prototype engine, equipped with the CR-FIS. In this case, however, the employment of a Wiebe equation was no longer permitted. In order to reproduce the effects of the injection parameters modulation, a direct modeling of the spray behavior, fuel-air mixing and combustion process was required. For this reason, inside the optimization process, the GT-Power built-in DIJet (Direct-Injection-Jet) model was utilized, as explained in the next sections.

cnr

Fig. 13 - Logic scheme of the optimization process within modeFRONTIER®

Combustion Noise Estimation
It is well known that the combustion noise generation mechanism is a complex phenomenon, including non-stationary and non-linear effects. In-cylinder pressure gradients during combustion process are considered the main excitation forces [20] on the cylinder liner and engine head. Additional contributions come, however, through the excitations exerted on the crankshaft by the inertia forces occurring as a consequence of the rotation and alternative motion of various engine components. While the first contribution depends on various operating parameters (engine speed, load, injection phasing and strategy, etc.), the second term is usually related solely to engine speed. Of course, along the sound propagation pattern from its generation inside the cylinder up to the noise acquisition location (usually at 1 meter from the engine block), the engine structure itself exerts a strong influence [21], in terms of natural vibration frequencies and vibration modes. The structure behavior is often synthesized in terms of a structural attenuation curve [16]. However, a more recent methodology was proposed [14] for the prediction of the overall combustion noise, which includes in the correlation a strict dependency on the engine operating conditions and injection strategy. The above described approach was used in this work. The main idea behind this technique was to decompose the total incylinder pressure signal (ptot) according to three main contributions: compression-expansion (pmot), combustion (pcomb) and resonance (pres) pressures:

cnr

The first contribution (also referred as pseudo-motored signal) is only related to volume variation, and was used as a reference signal. It was determined by a direct in-cylinder pressure acquisition during a fuel switch-off operation. The sum of combustion and resonance pressures is also referred to as excess pressure (pexcess) and was determined by the difference between the total and the pseudo-motored pressures:

cnr

This contribution is of course related to both fuel burning and to high-frequency resonant pressure fluctuations, induced by the pressure gradients during the combustion process [15]. To separate the above two terms, a high pass-band filter of the total pressure FFT was accomplished, as shown in figure 9. Above a proper cut-off frequency (about 4.5kHz) in fact, the pressure amplitudes (expressed in dB) tend to increase, thus indicating the occurrence of a resonance phenomenon [14]. An IFFT procedure was applied to the high pass-band filtered signal allowing to finally reconstruct the three contributions in eq. (1). They were compared together in figure 10. Despite the presence of the previously discussed high-frequency amplitudes, the resonant pressure was significantly lower than other contributions. Nevertheless, it may still exert a non-negligible effect on the overall noise. The three decomposed pressures were utilized to compute two characteristic indices I1 and I2 defined as:

cnr

cnr

n is the engine speed and nidle the idle rotational speed (fixed at 1000 rpm). The I1 index is a function of the maximum pressure gradient of the combustion contribution, occurring after the pilot (dpmax1/dt)comb and the main injection (dpmax2/dt)comb. The I1 index is also non-dimensionalized over the maximum pressure gradient of the pseudo-motored pressure (dpmax/dt)comb. In the case of a single-shot injection, a unique term is of course present in the eq. (3) numerator. The I2 index takes into account the acoustic energies (∫p2dt) associated with resonance and motored pressure signals. An additional index In is finally defined in [14], accounting for mechanical noise contribution, related, as stated, to the sole engine speed:

cnr

Basing on the above definitions, the Overall Noise (ON) can be finally computed as:

cnr

cnr

Fig. 14 - Scatter chart of the optimization process highlighting the Pareto Front

Ci being proper tuning constants, depending on the engine architecture and size. Following the relations (3-6), a Matlab routine was developed to properly process the in-cylinder pressure cycle and compute the various noise indices and the overall noise. This routine was applied to both the experimental and GT-Power computed pressure cycles. In the experimental analysis, the motored signal was directly acquired by means of a sudden fuel-switch off maneuver. Similarly, to compute the motored pressure, the fuel injection was completely disabled in the numerical analyses. Figure 11 compares the ON computed levels with the ones experimentally measured at the test-bench. Some adjustment of the Ci constants was required with respect to the values proposed in [14]. The agreement obtained through the employment of the experimental pressure cycles is satisfactory at each engine speed. A maximum absolute error of about 1.3 dB was found at a medium engine speed. As a consequence of the inaccuracies included in the engine simulation, the agreement at high speed slightly worsens when the predicted pressures are considered. Moreover, the employed zero dimensional model is unable to take into account the high frequency contributions of the computed pressure that are strictly related to the resonance phenomenon. Nevertheless, the satisfactory agreement shown authorizes the employment of the recalled methodology within the optimization procedure described in the next paragraph.

Optimization Procedure The previously described 1D and combustion noise models constituted the basis for the optimization of the injection strategy of the prototype CR engine. However, as already stated, a more advanced combustion model was in this case required. The latter (DIJet model) follows the multi-zone Hiroyasu approach [22-25] and is able to describe the fuel injection, break-up, airentrainment, evaporation and combustion processes. Details on the employed model can be found in [18,26]. The injector characteristics (in terms of holes number and diameter) injection strategy profile and timing represents the input data. The value of a number of tuning constants, acting as correction terms in the numerous correlations included in the model, were also assigned. Due to the absence of experimental data on the prototype engine, the tuning constants were identified in order to match the experimental pressure cycles measured on a similar engine, equipped with the same FIS, as reported in [26]. A pilot-plus-main strategy was specified for the CR-FIS at 3000 rpm, at full load conditions. Figure 12 shows the way the fuel injection strategy was schematized, based on the definition of three degrees of freedom, namely the start of pilot injection (SOIP), the dwell time between the pilot and main (DWELL) and the duration of the main injection (MDUR). In order to maintain the same injected fuel mass (22 mg), a constant overall duration (PDUR+MDUR=18.8°) was specified. Constant values of needlelift ramp-up (1.9°) and ramp-down (1.6°) were also assigned, depending on its dynamics. This allowed the computation of the crank angle position in all main points of the injection strategy, as a function of the 3 parameters SOIP, DWELL and MDUR:

cnr

The logical development of the optimization problem within the modeFRONTIER® environment is explained in figure 13. A number of Transfer Variables objects - together with the three Input Variables SOIP, DWELL and MDUR - were defined based on relations (7), and were written inside the GT-Power input file (LD500.dat). For each set of the above parameters, a proper script procedure runs the GT-Power code and extracts the incylinder pressure (pressure3000rpm), which is required by the Matlab routine computing the overall combustion noise. A multiobjective optimization was so defined to contemporarily search the maximum Indicated Mean Effective Pressure (IMEP) and the minimum of the overall noise. To solve the above problem, the MOGA-II algorithm was utilized, belonging to the category of genetic algorithms [27] and employing a range adaptation technique to carry out time-consuming evaluations. Figure 14 displays a scatter chart of the optimization procedure highlighting the Pareto Frontier occurring when the IMEP is plotted against the ON (a solution is said to be Pareto optimal if there is no other solution which is better in all objectives). A trade-off between the two objectives has clearly occurred. The position of the Base Engine, also shown in the same figure, was far away from the Pareto Frontier, thus indicating the possibility to attain a better level of both objectives. In order to select a single solution among the ones located on the Pareto frontiers, the “Multi Criteria Decision Making” tool (MCDM) provided in modeFRONTIER® was employed. It allows the definition of preferences expressed by the user through direct specification of attributes of importance (weights) among the various objectives. Depending on the above relations, the MCDM tool was able to classify all the Pareto Frontier solutions with a decreasing rank value.

Two different specifications were attempted: in the first case, the IMEP was considered the most important objective and assumed a weight two times higher than ON. Under this hypothesis, the point gaining the highest rank was the “MCDM Solution #1”, whose position is highlighted in figure 14. In this way both a ON reduction and a small IMEP increase was obtained with respect to the Base Engine. As an alternative, the same weight was specified for both IMEP and ON. In this other case, the solution #2 was selected and a small IMEP reduction was accepted to obtain a more relevant ON drop. The position of the two solutions puts into evidence that the MCDM procedure effectively realized a compromise between the conflicting needs, quantified by the attributes of importance previously described. In addition, this procedure defined a standardized method for the selection of the “global” optimum. Figure 15 and table 1 compare the optimal injection strategies selected by the MCDM tool and synthesize the related outputs. High IMEP required an advanced start of both pilot and main injections, with a reduced dwell time. As expected, a lower ON is indeed found with a delayed SOI and a higher dwell time. Figure 16 finally compares the base engine pressure cycle with the ones obtained in the CR engine in correspondence with the previously shown injection strategies. The retarded combustion process occurring in both the optimized cases contemporarily determined a lower pressure peak and a reduced pressure gradient (lower noise). Contemporary, a higher pressure level was found during the expansion stroke, which was mainly responsible of the small IMEP increase in the MCDM #1 solution.

cnr

Fig. 15 - Optimal injection strategies

cnr

Fig. 16 - Comparisons of the pressure cycles obtained through the optimization process

Conclusion
The paper described a methodology for the identification of optimal injection strategies of a CR light-duty diesel engine. The above objective was reached through the development of proper models for the prediction of engine performance and combustion noise. Both models were validated in reference to experimental data collected on a base, mechanical injection engine. Then a multi-objective optimization process was carried out with the aim of characterizing the trade-off between IMEP and ON on the prototype CR engine. A standardized procedure was also defined in order to select a unique solution, on the base of the user preferences and weight of importance of the single objective. The optimization procedure was able to capture the expected effects of the injection strategy parameters on the overall performance and radiated noise. It represents a very useful tool to reduce the huge experimental activity usually required to develop the control logic of the FIS. The methodology can be easily extended to multiple operating conditions and can include additional constraints related, for example, to noxious species emission predicted through 3D-CFD analyses.

Acknowledgements
The authors would like to express their thanks to Dr. Gerardo Valentino for supporting the experimental activity carried out in the present study.

 

Articolo pubblicato sulla Newsletter EnginSoft Anno 7 n°3

Daniela Siano (Researcher)
Istituto Motori CNR

Fabio Bozza (Full Professor)
Università di Napoli “Federico II”

 
EnginSoft SpA | © All rights reserved | VAT nb IT00599320223
Questo sito utilizza cookie tecnici e cookie analytics, che raccolgono dati in forma aggregata | Informativa completa sulla privacy e sui cookie