International journal of automotive technology, tập 10, số 2, 2009

131 − 139 2009131 COMBUSTION AND EMISSION CHARACTERISTICS OF BD20 REFORMED BY ULTRASONIC ENERGY FOR DIFFERENT INJECTION DELAY AND EGR RATE IN A DIESEL ENGINE S.. These adjustments norma

International Journal of Automotive Technology , Vol 10, No 2, pp 131 − 139 (2009)

131

COMBUSTION AND EMISSION CHARACTERISTICS OF BD20 REFORMED BY ULTRASONIC ENERGY FOR DIFFERENT INJECTION

DELAY AND EGR RATE IN A DIESEL ENGINE

S Y IM 1) , D S CHOI 2) and J I RYU 3)*

Daejeon 305-764, Korea

Daejeon 305-764, Korea

(Received 25 April 2007; Revised 11 September 2008)

ABSTRACT− The purpose of this study is to understand the operational characteristics of a diesel engine that uses BD20 reformed by ultrasonic energy irradiation In particular we study the effects of tuning injection delay and EGR rate BD containing about 10% oxygen has attracted attention due to soaring crude oil prices and environmental pollution This oxygen decreases soot by promoting combustion, but it also increases NOx To solve this problem, injection timing may be delayed

or an EGR system may be applied These adjustments normally lower engine power and increase exhaust emission but, in using fuel reformed by ultrasonic energy irradiation (which is changed physically and chemically to promote combustion), we may hope to circumvent this problem To control the duration of the ultrasonic energy irradiation, the capacity of the chamber

in an ultrasonic energy fuel supply system was tested at 550cc and 1100cc capacities As for the results of the experiment, we could identify the optimum EGR rate by investigating the engine performance and the characteristics of exhaust emissions according to the injection timing and the EGR rate while ultrasonically irradiated BD20 was fed to a commercial diesel engine With UBD20 (at an injection timing of BTDC 16 o ), the optimum EGR rate, giving satisfactory engine performance and exhaust emissions characteristics, was in the range of 15~20%.

KEY WORDS : BD (biodiesel fuel), BD20 (diesel 80% + biodiesel fuel 20%), EGR (exhaust gas recirculation), Injection delay, UBD (ultrasonic energy irradiated biodiesel fuel), Ultrasonic energy irradiation

1 INTRODUCTION

Due to soaring crude oil prices and environmental

pro-blems, interest in biodiesel fuel (hereafter called BD) has

grown BD is similar to diesel fuel and can be applied to

commercial diesel engines without any special modification

Although BD contains about 10% oxygen and thereby

decreases soot by promoting combustion, this active

com-bustion leads to an increase in NOx emissions due to the

Ryu and Oh, 2003; Baik, 2006; Bae et al., 2002; Oh et al.,

2002)

In particular, BD20, which is a blend of diesel fuel and

BD in a volume ratio of 8:2, contains a substantial quantity

To overcome this problem, the EGR (Exhaust Gas

Re-circulation) system has been widely put into practical use

The EGR system is inexpensive to install, but emission of

compounds other than NOx are known increase with anincrease of the EGR rate There are also a few problems tosolve in order for it to be put into practice universally,including contamination of the engine induction system byincrease of the break specific fuel consumption (BSFC)and soot, the abrasion and corrosion of parts in the engine,

Ham and Chun, 2002)

When fuel is irradiated by ultrasonic energy, the fuelundergoes physical and chemical changes induced by theenergy of the ultrasonic irradiation For example, thearomatic constituents become constituents of the fattygroup, and the isoparaffins turn into normal paraffins Thisleads to an increase in the cetane number and heatingvalue, while decreasing viscosity, surface tension, and thespray diameter (SMD) In this way, fuel quality is reformed,physically and chemically, promoting combustion and de-creasing soot (Choi, 1996; Jeong et al., 1991; Lee and Ryu,2003; Song, 2005)

There is a trade-off however, as NOx reduction leads tothe increase of soot, while soot reduction leads to the

*Corresponding author. e-mail: ryuji@cnu.ac.kr

132 S Y IM, D S CHOI and J I RYU

increase of NOx Hence, it is very difficult to reduce both

quantities simultaneously The fuel injection timing is

markedly delayed to reduce NOx, because this leads to a

raise in the temperature of combustion The temperature at

the intake also increases when using hot EGR

Alternative-ly, cooled EGR may be used to display a constant effect in

the range over mid-load The EGR method has been widely

applied to small diesel vehicles but the soot in the exhaust

gas contains sulfur oxides that enter the engine, causing

enhanced wear on the piston and cylinders The life of the

engine oil is also badly affected Hence it is essential to

designate the driving range with a large amount of NOx in

order to maintain an appropriate EGR rate (Uchida, 1993)

Therefore, this study aims to decide the optimum

injec-tion timing and the EGR rate to obtain a simultaneous

reduction of soot and NOx to within minimum values,

without deteriorating engine performance using BD20

reformed by ultrasonic energy irradiation

2 EXPERIMENTAL SYSTEM AND METHODS

2.1 Ultrasonic Energy Fuel Supply System

Figure 1 provides a schematic diagram for the ultrasonic

energy fuel supply system used for ultrasonic irradiation

The specifications of the transducer used in this experimentare presented in Table 1

A bolted Langevin transducer (BLT) was used for theultrasonic energy fuel supply system Its structure is suchthat when an ultrasonic oscillator (50W) functions at AC220V, ultrasonic vibrations are produced at 28 kHz and aretransmitted through the horn and into the fuel To maxi-mize the ultrasonic energy irradiation, fuel is supplied tothe lower part of the chamber of the ultrasonic energy fuelsupply system and flows out through the upper part An airvent was installed in the back of the upper reflector of thechamber so that the air bubbles generated by the ultrasoniccavitations could escape (Dale Ensminger, 1988) The ultra-sonic energy fuel supply system was connected between afuel flowmeter and an injection pump so that the reformed(irradiated) fuel could be provided to the engine.2.2 Experimental System and Method

The schematic diagram and photograph of the mental apparatus is presented in Figures 2 and Figure 3.The apparatus consists of a dynamometer, a test engine,measurement instruments, and a data acquisition system.The specifications of the test engine are shown in Table 2.This experiment was performed according to the KSRISO

experi-Figure 1 Schematic diagram of an ultrasonic energy fuel

Figure 2 Schematic diagram of experimental apparatus

Figure 3 Photograph of experimental apparatus

COMBUSTION AND EMISSION CHARACTERISTICS OF BD20 REFORMED BY ULTRASONIC ENERGY 133

2534 testing standard (KSRISO, 2003) The engine load is

adjusted after fixing the engine speed Also, the injection

timing was set for BTDC 11o and 16o, with the engine load

set to 25%, 50%, 75% and 100% according to a maximum

torque for each engine speed The engine speed was varied

from 1000 rpm to 3500 rpm in 500 rpm intervals

Table 3 shows properties of the test fuels used for the

experiment BD is made out of soybean oil and in our tests

was blended with 20% (by volumetric ratio) commercial

diesel fuel

In order to investigate the effects of the duration of the

ultrasonic energy irradiation, the capacity of the chamber

for the ultrasonic energy fuel supply system was set to

either 500cc or 1100cc If the capacity of the chamber is

too small, the influence of the ultrasonic energy irradiation

is not significant, and if the capacity of the chamber is too

large, the temperature of the supplied fuel becomes high

enough to negatively affect the engine The amount of

recirculating exhaust gas is controlled by the EGR valve

was computed using the following formula (1):

(1)

Here, V 0 is intake air volume (m3/h) with no EGR, while V a

is that when EGR is taken into account

Since the temperature of the EGR gas changes in ance with each engine load, it was controlled here to be

used to remove particles in the recirculated exhaust gas.The temperature of the cooling water was maintained at

After the completion of each experiment, the fuel filter wasreplaced, and the fuel supply system was examined forneed of repair to ensure that previous experiments wouldnot affect subsequent ones

The list of equipment used for testing engine mance, combustion characteristics, and exhaust emissions

perfor-is presented in Table 4

3 CHARACTERISTICS OF INJECTION DELAY

Figure 4 shows the relationship between the engine powerand the engine speed for the test fuels (commercial diesel,BD20) according to the amount of ultrasonic energy irradi-ation Here, the chamber capacities were 550cc and 1100cc

diesel fuel was 0~3% higher than that for BD20 Withregard to total engine speed, UBD20 was 0~2% higher than

EGR rate % ( )V 0 – V a

V 0 -× 100

Table 2 Specifications of test engine

4 cylinder diesel engineCombustion chamber type Swirl combustion chamber

(Over Head Cam shaft)

Table 3 Properties of test fuels

A/D-D/A converter National Ins Co type PCI-6024E

Fuel flow meter AND Co type HF-2000GDCapacity: 2100g,

Resolution: 0.01gSmoke meter Su Kwang precision Co type Gold 707Measurable range: 0~100%±2% F.SNOx analyzer HORIBA KOREA Ltd MEXA -554JKNOX Chemical methodMeasurable range: 0~5,000 ppm

±20 ppm

CO analyzer BOSCH ETT 008.55 Non-Dispersive Infrared MethodMeasurable range :

0.00~10.00% vol ±0.06% vol

134 S Y IM, D S CHOI and J I RYU

diesel with a chamber capacity of 550cc, and 1~6% higher

than diesel with a chamber capacity of 1100cc In other

words, the engine power was enhanced the most when

using irradiated fuel For the BTDC 11o case, when the fuel

injection timing was delayed and ultrasonic irradiation was

applied, the engine power was enhanced or remained

almost the same within a 2~3% margin as compared to

BD20 over the entire range of speeds The fuel injection

timing was delayed to ensure optimum fuel injection

tim-ing for NOx reduction Although this delay has the effect of

reducing engine power, the fuel reformation effect due to

ultrasonic irradiation caused a compensating increase in

power Hence, the overall engine power observed for

UBD20 was similar to that for BD20 In fact, the engine

power characteristics observed for BD20, UBD20 and

diesel were similar throughout the whole range of engine

speeds used here With a fuel injection timing of BTDC

tendency toward power increase This implies that the

heating value of BD is lower than that of diesel, but that

fuel reformation, achieved by ultrasonic energy irradiation

lead to increased thermal efficiency

Figure 5 shows the maximum combustion pressure for

cases with and without ultrasonic irradiation with fuel

increas-ed by 5% using the 550cc chamber, and by approximately

2~6% with an 1100cc chamber

In addition, the UBD20 maximum combustion pressure

was found to be larger in the 550cc case by approximately

timing was delayed compared to that used for BD20)

This is because the fuel reformed by the ultrasonic

irradiation promotes combustion via an improvement in

ignition This is indicative of promoted evaporation of fuel

and reduced Sauter mean diameter (SMD) of droplets as

compared to the case of BD20

As stated above, the optimum fuel injection timing for

reduction of NOx in UBD20 is more delayed than that forBD20, and thus the maximum combustion pressure islower However, as presented in Figure 5, increasedduration of irradiation leads to somewhat of an increase inthe maximum combustion pressure Accordingly, to ensureoptimum fuel injection timing for UBD20, the maximumcombustion pressure was reduced slightly by delaying thefuel injection timing However, providing enough duration

of ultrasonic energy irradiation contributed to an effectivefuel reformation, which improved combustion The improve-ment seems to be due to the fact that the maximumcombustion pressure was similar to that for BD20 at BTDC

16o.Figure 6 compares the combustion pressure, the heatrelease rate and the mass burning rate of UBD20 with

was held at 75%, and the engine speed at 2,000 rpm.When UBD20 was supplied to the engine, its com-bustion started slightly later than that for BD20, although itwas found to have similar characteristics essentially, whenultrasonic energy irradiation was performed on BD20 (used

duration were shortened due to acceleration of combustion

It is thought because the 10% oxygen content of BDactively promotes combustion It is also thought that theultrasonic energy reforms fuel quality physically and chemi-cally leading to an increase in heating value and ignitionquality, while decreasing the viscosity, the surface tensionand the spray diameter (SMD) size

Figure 7 shows the brake specific fuel consumption(BSFC) of the engine according to the duration of ultra-sonic energy irradiation (as implied by the chamber capa-city) at fuel injection timings of BTDC 11o and BTDC 16o

1~2% higher than that of BD20 Over the whole range ofengine speed, the BSFC of UBD20 was 1~2% lower with a550cc chamber, and 1~3% lower with a 1100cc chamber.Thus, the BSFC was most enhanced for the irradiated

Figure 4 Comparison of engine power under varying

engine speed at engine load 75%

Figure 5 Comparison of Pmax under varying load atengine speed of 2,000 rpm

COMBUSTION AND EMISSION CHARACTERISTICS OF BD20 REFORMED BY ULTRASONIC ENERGY 135

fuel With delayed timing, UBD20 showed a BSFC that

was better than the commercial diesel by 1~2% over the

whole range of engine speeds

As mentioned above, it was anticipated that BSFC

would be deteriorated by delaying the fuel injection timing

in order to obtain optimal NOx reduction for UBD20 In

reality, for fuel reformed by sufficient ultrasonic energy

irradiation, the overall BSFC was enhanced Furthermore,

BSFC gradually decreased with the amount of irradiation

of BD, but not when such a procedure was applied to

commercial diesel It is believed that the oxygen contained

in BD promotes combustion, which then leads to enhanced

combustion efficiency

Figure 8 shows NOx characteristics according to engine

speed and the duration of ultrasonic energy irradiation

given to the test fuels (commercial diesel, BD20, where

duration is implied by chamber capacity)

The engine load was 75% and the fuel injection timing

emission from commercial diesel fuel was 3~5% less thanthat of BD20 over the whole range of engine speeds.UBD20 NOx was 4~26% higher than BD20 with a 550ccchamber, and 25~42% higher with an 1100cc chamber

Figure 6 Comparison of cylinder pressure, heat release rate

and mass burning rate at an engine speed of 2,000 rpm and

136 S Y IM, D S CHOI and J I RYU

As mentioned in the discussion of engine performance,

the ultrasonic energy irradiation reformed the fuel quality

physically and chemically, which promoted combustion

This leads to a pressure increase in the combustion

cham-ber, and a resultant increase in combustion temperature

Higher temperature leads to increased NOx emissions as

compared to BD20

NOx tends to generally increase with BD content,

because oxygen contained in BD promotes combustion,

thereby raising the combustion chamber temperature When

the fuel was irradiated, NOx was remarkably increased as a

result of the reformed fuel

Figure 9 shows the soot characteristics according to

engine speed depending on the duration of irradiation

(implied by chamber capacity) The engine load was 75%

The soot of commercial diesel fuel was 13~60% higher

than that from BD20 over the entire speed range For

UBD20, it was 13~33% lower with a 550cc chamber and

40~67% lower an 1100cc chamber Notably, when the fuel

the ultrasonic energy irradiation, the soot level was 20~

60% lower than that for BD20 over the entire speed range

Although the combustion duration was not enough (due to

the delayed timing), the soot was reduced because of the

physical and chemical properties of irradiated fuel

4 CHARACTERISTICS RELATED TO

CHANGES OF EGR RATE

Figure 10 shows the maximum combustion pressure

the maximum pressure with the EGR rate showed a

minimum of 1.4% at an EGR of 20%, while showing a

dramatic decrease of 5% at an EGR rate of over 30%

It is thought that, at an EGR rate of over 30%, the

maximum combustion pressure quickly decreases because

the concentration of oxygen becomes insufficient for

combustion

Figure 11 shows the relationship between the enginepower and the EGR rate for the test fuels (commercialdiesel, BD20) according to irradiation when the enginespeed is 2000 rpm and engine load is 75% with fuel

fuel was 4~11% higher than that for BD20 For UBD20,the engine power was 6~11% higher with an 1100cc cham-ber, and 4~5% higher with a 550cc chamber For BTDC

EGR was applied, the engine power of BD20 tended todecrease more than that of the commercial diesel fuel Withincreased irradiation, the engine power increased, similar

to the non-EGR case

Figure 12 shows the relationship between the BSFC andthe EGR rate of test fuels (commercial diesel, BD20)according to irradiation with an engine speed of 2000 rpm

commercial diesel fuel was larger by 2~4% than that ofBD20 The BSFC for UBD20 was about 3~5% lower than

Figure 10 Comparison of Pmax under varying EGR Rate

COMBUSTION AND EMISSION CHARACTERISTICS OF BD20 REFORMED BY ULTRASONIC ENERGY 137

that for BD10 while using an 1100cc chamber, and about

2% lower while using a 550cc chamber At BTDC 11~ the

BSFC for UBD20 was about 3% lower

Varying the fuel injection timing at an EGR rate of 10%

does not make a significant difference in the BSFC under

identical engine speed and load However, the range of

fluctuation tended to expand with increasing EGR rate

Figure 13 shows the relationship of NOx according to

the EGR rate and irradiation for the test fuels (commercial

diesel, BD20) at an engine speed of 2000 rpm and load of

NOx from commercial diesel fuel was 2~8% lower than

that of BD20 UBD20 NOx was 5~8% higher than BD20

with a 550cc chamber, and 8~20% higher with an 1100cc

chamber

to BD20 over the whole range of engine speed Against this

backdrop, it was reassuring that reduction of NOx was

greatly affected by the fuel injection timing

NOx was dramatically decreased in accordance with the

increase of the EGR rate, especially in the 40% range

where the reduction was most notable When EGR was

applied to the engine, some of the air inhaled into the

combustion cylinder was replaced with inert exhaust gas

Because of this combustion temperature was lowered and

combustion was delayed, eventually leading to a

remark-able reduction of NOx NOx emissions were, therefore,

rapidly reduced with an increase of the EGR rate In

addition, the NOx reduction rate was found to increase

with increasing engine load, but it could be predicted that

exhaust emissions would be remarkably increased due to

the decrease of oxygen

Figure 14 shows the relationship between soot and the

EGR rate of the test fuels (commercial diesel, BD20)

according to irradiation at an engine speed of 2000 rpm,

When the fuel injection timing was delayed togetherwith application of ultrasonic energy irradiation, soot wasdecreased by 34~43% over the entire range when com-pared to the soot from commercial diesel fuel Althoughthe combustion duration was insufficient due to the delayedfuel injection timing, the soot was reduced because ofvarious factors related to the ultrasonic energy that promotecombustion In other words, soot increased with an increase

in EGR rate, but the case of mixed diesel/BD emitted lesssoot than with diesel alone

When the EGR rate was more than 20%, soot wasincreased over the whole range The recirculated exhaustgas reduced the amount of oxygen in the intake air that wasinhaled into the combustion chamber, disturbing the com-bustion process

In addition, when the EGR was applied to BD20, sootwas reduced by a larger amount than was seen with dieselfuel alone This is because oxygen contained in BD pro-moted oxidation of the fuel It is thought that oxygencontained in the fuel promoted combustion by promptingchemical reactions with hydrocarbons Exceptional sootreduction was achieved over the whole range of the EGRrate due to the irradiation This is because the ultrasonicfuel reformation, together with biodiesel fuel elements,contributed to combustion promotion

Figures 15 and 16 show the correlation of the BSFC,NOx, engine power and soot with cooled EGR rate at anengine speed of 2000 rpm and load of 75% for BD20 at

11o.The BSFC and soot for both BD20 and the reformed fuel(UBD20) increased in accordance with the increase of theEGR rate, while engine power and NOx tended to gradu-ally decrease

Soot increased with the increase in the EGR rate, but itwas emitted relatively less with BD20 than with commer-cial diesel fuel Soot was also emitted relatively less fromthe irradiated fuel, UBD20

However, with an EGR rate of more than 20%, the soot

Figure 13 Comparison of NOx under varying the EGR

Rates at engine load 75% (2000 rpm)

Figure 14 Comparison of soot under varying EGR Rates at

an engine load 75% (2000 rpm)

138 S Y IM, D S CHOI and J I RYU

rapidly increased This is because the recirculated exhaust

gas reduces the amount of oxygen sucked into the

bustion chamber, resulting in insufficient oxygen for

com-bustion

In addition, NOx was dramatically reduced in

accord-ance with the increase in EGR rate As described

previ-ously, this feature of the EGR is due to lower combustion

temperature and slackened combustion speed

rate to satisfy the BSFC, NOx, engine power and soot

considerations simultaneously is in the range of 10~20%

soot levels were similar or a little more than in the case ofthe commercial diesel fuel or BD20 If the EGR rate wasless than 5%, NOx levels were similar to the case ofcommercial diesel fuel

Therefore, when UBD20 was used at the fuel injection

15~20% to reduce both soot and NOx Against thisbackdrop, the optimum EGR rate should be considered inthe range that does not deteriorate characteristics of theengine performance and minimizes exhaust emissions

5 CONCLUSION

We studied the use of fuel reformed by ultrasonic energyirradiation in diesel engines in the context of optimizingperformance and emissions In particular, we studied thetuning of fuel injection timing and EGR rate to obtain thefollowing results:

(1) The maximum combustion pressure of the chamberincreased by up to 6% with an engine speed of 2,000rpm and load of 75% upon irradiating BD20 When

the pressure increased by up to 3%

(2) For the reduction of NOx from biodiesel fuel (anoxygenated fuel with high NO), the optimum injectiontiming of the fuel reformed by ultrasonic energy irradi-ation should be delayed compared to that of commonlyused diesel fuel As regards BSFC, the results werefound to improve as the ultrasonic energy irradiationduration became longer (the chamber capacity larger).(3) NOx emission from UBD20 was 42% higher thanBD20 over the whole range of this experiment, whilesoot was a maximum of 67% lower

(4) When BD20 was used with a fuel injection timing of

satisfac-tory BSFC and engine power was in the range of10~20%

(5) When BD20 was used with the fuel injection timing of

and NOx was in the range of 15~20%

When the fuel reformed by the ultrasonic energy irradiationwas applied to the diesel engine, the optimum EGR ratewas identified to be 15~20% to reduce NOx and topromote the BSFC Also, when the fuel injection timingwas delayed and the duration of the ultrasonic energyirradiation was prolonged, the reduction effect increased.For reformed fuel and delayed timing (UBD20 with BTDC

and NOx were both reduced as compared to BD20 levels

We find that in order to enhance engine performance andreduce exhaust emissions it is essential to precisely controlthe injection timing and the duration of ultrasonic energyirradiation and to consider the optimum EGR rate

Figure 15 Comparison of NOx vs BSFC under varying

the EGR Rates at engine load of 75% (2000 rpm)

Figure 16 Comparison of Power vs Soot under varying

the EGR Rates at engine load of 75% (2000 rpm)

COMBUSTION AND EMISSION CHARACTERISTICS OF BD20 REFORMED BY ULTRASONIC ENERGY 139

REFERENCES

Agarwal, D., Sinha, S and Agarwal, A K (2006)

Experi-mental investigation of control of NOx emissions in

Energy 31, 14, 2356−2369

Bae, M W., Ryu, C S., Kim, H M., Tsuchiya, K and

Mochimaru, Y (2002) Effect of intake mixture

temper-ature on performance and exhaust emissions in diesel

68th Annual Conf Japan Institution of Marine

Engi-neering, 51−58

Baik, D S (2006) Combined effects of BD20, low sulfur

diesel fuel and diesel oxidation catalyst in a HD diesel

engine Int J Automotive Technology 7, 6, 653−658

Charac-teristics and the Engine Performance CharacCharac-teristics of

the Ultrasonic-Energy-Added Diesel Fuel Ph.D

Disser-tation Chungnam University Daejeon Korea

edn New York

Ham, Y Y and Chun, K M (2002) Parametric study for

reducing NO and soot emissions in a DI diesel engine by

Automotive Engineers 10, 5, 35−44

Jeong, M J., Cho, G S and Ryu, J I (1991) The

per-formance imorovement of the diesel engine by the

ultrasonic vibrations Trans Korean Society of Automotive

Engineers 13, 6, 65−71

-Gross Shaft Power, Korean Agency for Technology and Standards. Gyeonggi Korea

Lee, B H and Ryu, J I (2003) A study on relationshipbetween fuel characteristics and combustion characteri-stics of reformed diesel fuels by ultrasonic energy irradi-ation(II)-Relationship between chemical structure and

Engineers 11, 1, 64−71

Lue, Y F., Yeh, Y Y and Wu, C H (2001) The emissioncharacteristics of a small D I diesel engine using bio-diesel blended fuels J Environmental Science and Health

and Exhaust in Diesel Engine of Ultrasonic Irradiation Bio-diesel Blended Fuel Ph.D Dissertation ChungnamUniversity Daejeon Korea

Uchida, N (1993) Combined effects of EGR and

No 930601

International Journal of Automotive Technology , Vol 10, No 2, pp 141 − 149 (2009)

141

STRATEGIES FOR IMPROVING THE MODE TRANSITION IN A SEQUENTIAL PARALLEL TURBOCHARGED AUTOMOTIVE DIESEL

ENGINE

J GALINDO, H CLIMENT, C GUARDIOLA * and J DOMÉNECH

CMT Motores Térmicos, Universidad Politécnica de Valencia, Camino de Vera s/n, Valencia 46022, Spain

(Received 1 August 2007; Revised 2 September 2008)

ABSTRACT− Parallel sequential turbocharging systems are able to operate in different modes, which are defined according

to the turbochargers that simultaneously boost the engine, and are controlled by means of specific valves In order to cover the full engine operating range, a smooth transition between turbocharging operating modes must be ensured However, important disturbances affect both boost and exhaust pressure when shifting the operation mode, thus causing non-negligible torque oscillations This paper presents different methods for smoothing such undesirable effects during mode transition Strategies covering optimal synchronization of the control valves, control of the valves’ position, and correction of the injected fuel during the transition are analysed A fully instrumented passenger car engine is used for illustrating the different torque smoothing methods, and experimental results for transitions during both steady operation and engine accelerations are shown.

KEY WORDS : Sequential turbocharging, Diesel engine, Transition, Boost control

NOMEMCLATURE

Vrecirc: valve of the recirculation circuit

1 INTRODUCTION

Turbocharging is a widely used method that improves

internal combustion engine performance However,

turbo-charged engines cause a delay during load transients (Watson

and Janota, 1982)

The causes of the delay are mechanical, thermo- and

fluid dynamic The first involves the inertia of the

turbo-charger, the second includes the processes of mass and

energy transfer between the exhaust valve and the turbine,

al., 2002; Payri et al., 2002)

The problem of the turbo-lag is of great importancewhen an acceleration is demanded and exhaust gas energyavailability is low, which occurs at low engine speed andload In this case, the driver’s torque demand produces astep increase in the fuel supplied; however, the required air

to ensure a correct combustion process is not supplied bythe compressor as fast as it should be This situation results

in an increase in the fuel-to-air ratio, and hence, teristic ‘black smoke’ appears In order to limit smokeemissions, current diesel engines limit the injected fuelduring the accelerations on the basis of minimum air-to-fuel ratio criteria, which also negatively affects engine per-formance during load transients

charac-Current demands on diesel engine performance requirethe solving of problems related to the turbo-lag Downsiz-ing is a general trend that leads to engines that have thesame power output with lower displacement The down-sized engine has lower emissions and consumption but thelow end torque is worsened As a result, some new techno-logies have been developed, such as the variable geometry

2002), which are able to adapt the effective area of thestator Another method to improve the performance intransient conditions is to employ a smaller turbochargersince: a) the turbine is better designed for working with lowexhaust gases mass flow, and b) it accelerates faster because

of its lower inertia Both methods allow faster transientengine response However, small size turbochargers limitthe maximum engine speed and load In order to overcomethis problem, sequential operation of turbochargers arrang-

*Corresponding author. e-mail: carguaga@mot.upv.es

142 J GALINDO, H CLIMENT, C GUARDIOLA and J DOMÉNECH

ed in parallel can be used

Sequential parallel turbocharging systems are able to

improve diesel engine transient response, since low inertia

2002) This turbocharging system has been used in the past

for marine Diesel (Ren et al., 1998; Benvenuto and Campora,

2002; Kench and Klotz, 2002) and gasoline automotive

focused on the development of a passenger-car diesel

engine In contrast to marine applications, engine speed in

land vehicles is subject to more variation, and thus, the

dynamic torque response is a key factor In contrast to

conventional turbocharging systems, parallel sequential

systems are more complex and include non-standard

ele-ments Additional details on the advantages and

disadvant-ages of this kind of turbocharging system compared to

presented by Portalier et al (2006) and Galindo et al (2009)

Most research on diesel engines has been centered on

marine applications where the engine running conditions

are fairly constant; very few investigations have analyzed

the load transient that appears during the sequence from

one to two turbochargers in operation and vice versa This

process will be referred to hereafter as a transition

This paper presents experimental work related to a

sequential parallel double turbocharged diesel engine Since

the engine is designed for automotive applications,

impor-tant variations on both engine speed and load are expected

Thus, the main objective of the paper is to illustrate the

main phenomena causing torque oscillations during the

transition, and to present several strategies in order to

reduce them

2 TURBOCHARGING SYSTEM DESCRIPTION

In this section the sequential parallel turbocharging system

is described Parallel sequential turbocharging systems

com-prise two independent turbochargers (TC1 and TC2) One

of them (TC1) is always in operation, while the second one

(TC2) operates only in the high speed region of the engine

map

TC1 is a basic charger and operates throughout the

complete operational range However, it is smaller than the

equivalent turbocharger used in a conventional system, and

is therefore lighter and has lower inertia These

charac-teristics allow TC1 to accelerate faster in lower ranges of

speed and load where TC2 is not operating In addition, and

due to the smaller compressor size, higher boost pressures

are admissible at low engine speeds, as the surge limit of

TC1’s compressor is less restrictive than those of

conv-entional compressors

TC2 is even smaller than TC1 and operates at high

engine speed and load range, where both turbochargers are

working simultaneously The activation of TC2 is possible

by means of auxiliary air valves in the engine

The system is completed by additional valves that allow

managing the transition between the two basic operationmodes (1T and 2T), and controlling the boost pressure Aschematic of the system is shown in Figure 1, where the topplot shows valve positions in 1T operation mode and thebottom plot shows valve positions in 2T operation mode Inboth cases a waste-gate (WG) is used for controlling boostpressure A short description of the functionality of eachvalve is provided below:

(1) A waste-gate (WG) is used for boost pressure control.Although variable geometry turbines (VGT) could also

be used, this possibility was not considered due tooverall cost restrictions of the turbocharging system.WG-based boost control in the parallel sequential turbo-charging system is similar to conventional turbocharg-

ed engines Hence, only WG influence on the transitionwill be highlighted

(2) A control valve at turbine 2 (VT2) is used to feedturbine 2 (T2) when necessary This valve activates thetransition between 1T and 2T modes In addition, smallopenings of the VT2 are used to prepare the transitionfrom 1T mode to 2T mode

(3) A valve placed downstream of compressor 2 (VC2) isused to prevent internal air leaks from the intake mani-fold to C2 when this compressor is not operating BothVC2 and VT2 determine the turbocharging systemconfiguration The engine runs with two turbochargers

in parallel if both valves are open

(4) As previously mentioned, TC2 must never be

complete-ly stopped: lubrication and a minimum thermal levelmust be ensured in order to avoid breaking TC2 whenshifting from 1T to 2T operation A minimum flowthrough C2 is ensured by slightly opening VT2, andusing a recirculation circuit with a control valve (Vrecirc)

Figure 1 System schematic and valve configuration for thetwo main operation modes (top: 1T; bottom: 2T)

STRATEGIES FOR IMPROVING THE MODE TRANSITION IN A SEQUENTIAL PARALLEL TURBOCHARGED 143

This valve must be open in 1T mode and closed in 2T

mode to avoid air leaks in the intake manifold Beyond

ensuring lubrication and minimum temperature, the

recirculation circuit helps prepare the transition form

1T to 2T via the progressive acceleration of TC2

(5) Finally, the engine is also fitted with an EGR circuit

with an EGR control valve As this system only opens

in the 1T operation zone, no real interaction occurs with

the parallel sequential turbocharging system For clarity

reasons, all diagrams presented have been simplified

by excluding the EGR system

2.1 Operating Range Limits

Important limitations in the operation ranges of each of the

turbocharging modes (1T and 2T) exist, hence it is not

possible to run the engine in a specific turbocharger

confi-guration for the entire operation range A short theoretical

approach based on engine volumetric efficiency is now

discussed

Volumetric efficiency is defined as follows:

(1)

cylinders, V D is the cylinder displacement, ρ i is the intake

manifold air density, n is the engine speed, and i equals 0.5

in a four-stroke engine Equation (1) can be conveniently

transformed into:

(2)where the first term is equal to the compression ratio in the

compressor if the pressure losses in the air filter and in the

charge air cooler are neglected

Equation (2) expresses the linear relationship between

the air mass flow and the compression ratio for a constant

engine speed if it is assumed that the temperature at the

charge air cooler outlet and the volumetric efficiency remain

constant The equation is valid when no EGR is performed

and in steady operation, where there are no mass storage

effects (i.e variation of manifold pressure)

Equation (2) is used to plot three engine characteristic

lines for three different engine speeds in a typical

com-pressor chart and is shown in Figure 2 At low engine

speed, the compression ratio is limited by either the

energy availability At high engine speed, the compression

ratio, and hence the intake pressure, is limited by the

compressor over-speed zone

The design steps in conventional turbocharging systems

consists in selecting a single turbocompressor to cover the

full engine operation range Big turbochargers are needed

to cover the high speed region, and surge is usually a

limiting factor in the low speed range

A smaller turbocharger can be selected in sequential

parallel systems This turbocharger is better adapted to the

low speed region, but over-speed is a problem for highboost pressures and engine speeds

When the two turbochargers of the sequential systemoperate simultaneously in the 2T mode, equation (2)becomes:

(3)This means that the equivalent operation point is shifted

to the left on the compressor chart as the total flow issupplied by the two turbochargers In this case, the over-speed limit is not reached, but surges are possible at lowengine speeds and high boost pressures In addition, lowexhaust gas energy availability in this region makes 2Toperation impossible

These two limits are represented on the engine map inFigure 3 In addition to these two restrictions, some mildcriteria for the operating mode selection can be added: atlow engine load and speed, the engine transient response isfaster if running at 1T mode, so the 2T mode is notrecommended, and there is also a zone at medium load andhigh engine speed that corresponds to highway drivingconditions, where operation in 1T mode would result in

Figure 2 Engine operation points plotted on a compressorchart

Figure 3 Engine map illustrating operation mode tions

restric-144 J GALINDO, H CLIMENT, C GUARDIOLA and J DOMÉNECH

higher fuel consumption

3 EXPERIMENTAL SET-UP

The engine tested was a 4-stroke, 2.2 liter, four cylinder

diesel engine with two turbochargers arranged in parallel

The test bench configuration is shown in Figure 4 The

engine is controlled by an ECU that is externally calibrated

by means of an ETAS ES1000 system In addition, a signal

bypass is used to include the sequential parallel

turbo-charging management system: an independent module for

controlling the four listed valves was programmed and

integrated into the system using ASCET software and an

ES1130 card A bypass to the ECU fueling strategy was

also available, which enabled fuel injected quantity

correc-tions during the transition

An eddy current dynamometer was used to load the

engine and perform transient tests ECU and ES1130 card

variables were acquired using INCA software In addition,

intake and exhaust manifold pressure and temperature,

engine speed and torque, and turbocharger speeds were

acquired with a PUMA system Instantaneous opacity

mea-surements were performed with an AVL439 opacimeter

4 TRANSITION AND TORQUE OSCILLATIONS

Although the sequential parallel turbocharging system in

diesel engines presents several advantages as previously

described, there is a significant drawback: the transition

between both operation modes becomes critical Torque

oscillations during the transition are of little importance in

marine applications, where these types of engines have

been largely employed, since the engines run in nearly

steady conditions with few transitions However, the

pro-blem is notable in automotive engines due to the higher

potential negative effects of torque variations on the vehicle

driveability

Several dynamic processes occur when the transitiontakes place, including variations in both intake and exhaustmanifold pressure, acceleration of the turbochargers, andthermal transients All these phenomena interact with thecombustion process and the engine pumping work, thusresulting in a non-steady evolution of engine torque duringthe transition

Engine efficiency can vary depending on the operationmode and during the transition The difference in pumpinglosses and the variation in combustion efficiency, whichstrongly depends on the in-cylinder pressure at the end ofthe intake process, are primary factors affecting engineefficiency Thus, different steady torque values are obtain-

ed when operating with the same fuel mass in 1T and 2Toperation modes, and oscillations appear in the torqueprofile

In addition, the sensed air mass flow during the tion can undergo important oscillations, which can triggerthe injection fuel limitation process to control smoke Inorder to avoid the interaction of this strategy with torqueoscillation due exclusively to the transition process, thesmoke control strategy was deactivated during the transi-tion There were two reasons underlying this decision:(1) The smoke control strategy is based on the measure-ment of a hot wire anemometer placed just downstream

transi-of the air filter During the transition from 2T to 1Toperation, the pressure of the portion of intake mani-fold between C2 and VC2 decreases This rapid varia-tion in the mass stored in the manifold causes anevident variation in the sensed air mass flow, while theadmitted air mass flow is less affected Consequently,the smoke control methodology overestimates therequired fuelling restriction and causes an additionaltorque drop

(2) The opacity level during the transition becomes important when compared with the beginning of a loadtransient Smoke emission during the transition is ashort-term smoke peak that could presumably be dealtwith via other emission control techniques like parti-culate filters This decision can be made during thefinal calibration

un-Hence, none of the experimental results in the presentpaper consider the fuelling limitation due to the smokecontrol strategy However, opacity measurements areprovided in order to check the transition impact on smokeemissions

Additionally, valve management during the transition iscritical, and slight variations in the timing of the closingand opening process of different involved valves can result

in magnification of the torque oscillation Controlling thisbecomes a critical issue for this type of engine (Cantemir,2001)

To achieve optimal valve synchronisation requires aniterative optimization process In this case a pilot study was

with experimental tests used later for fine tuning and

Figure 4 Scheme of the experimental setup Main

infor-mation fluxes are marked as gray arrows, while main

control actions are marked as black arrows

STRATEGIES FOR IMPROVING THE MODE TRANSITION IN A SEQUENTIAL PARALLEL TURBOCHARGED 145

validation The following paragraphs detail the transition

from1T to 2T and from 2T to 1T

4.1 1T to 2T Transition

When passing from 1T to 2T, C2 must be accelerated and

the part of the intake system between C2 and VC2 must be

pressurized Hence, additional energy is needed to perform

these actions

A first step when performing a transition is to close the

waste gate to capture all available exhaust gas energy The

next step is opening VT2 to enable exhaust gases to flow

through turbine 2 and accelerate the turbocharger All these

steps are done with VC2 closed and Vrecirc open in order

to avoid surge and reverse flow through the compressor

Once the C2 outlet pressure is close to the intake manifold

pressure, VC2 opens and Vrecirc closes

The top plots in Figure 5 depict the torque evolution

during the transition from mode 1T to 2T; the bottom plots

illustrate the valve opening sequence applied The

left-hand plots show the transition at 3250 rpm and 10 mg of

fuel per stroke, and the right hand plots show a similar

transition at the same speed and 30 mg of fuel per stroke It

is notable that the engine torque under steady conditions in

1T and 2T modes is not the same, since these engine tests

were performed with constant injected fuel

To illustrate the sensibility of the system to WG

actua-tion timing, two tests are shown in the figure: one using

optimal timing, and the other increasing the WG closing by

400 ms When the WG closes too early there is an increase

in the exhaust manifold pressure which increases the

pumping losses and reduces the engine torque (Luján et al.,

2007) It is possible to achieve a synchronization between

the WG and the VT2 in order to obtain a smooth torque

evolution This synchronization is more important at low

rather than at high engine loads The peak value of the

torque deterioration is 2 Nm for the low load case and 4

Nm for the high load case, but the relative variation is 8%

and 1.5%, respectively

Vrecirc and VC2 timings are not shown as they were notimportant to the transition: recirculation is used for pre-serving C2 from surge, and has no direct effect on thetorque evolution On the other hand, VC2 pneumaticactuation system does not enough force to open the valve inthe case of reverse pressure gradient Thus, once thecontrol action of VC2 is set to ‘open’, VC2 acts as a semi-passive valve, which automatically opens when pressure atC2 outlet is about equal to the intake manifold pressure.4.2 2T to 1T Transition

Although less critical, valve synchronization during thetransition from the 2T operating mode to 1T is alsorelevant In this case, the WG must open at the same timeVT2 closes A delay in opening the waste gate generates ahigher torque drop, due to the increase in backpressure andtherefore, the pumping losses increase

Figure 6 illustrates transitions from the 2T to 1T mode at

3250 rpm and 10 and 30 mg/stroke In this case, the delaybetween the two tests performed at the same enginecondition is 200 ms However, the sensitivity to WG timing

is not as high as in the 1T to 2T transition

It is noticeable that the torque dynamics during the 2T to1T transition are faster than in the 1T to 2T case, and asteady value is reached in less than 1 s

5 TRANSITION IMPROVEMENT STRATEGIES

Basic transitions shown in the previous section have, even

in the optimal synchronization case, two undesirable effects

on the torque: the first is that the different modes haveimportant differences in steady operation torque, and thesecond is oscillations in torque, especially in the 1T to 2Ttransition

This section is focused on different strategies for ing the torque evolution For steady operation torque differ-ence, it is necessary to correct the injected fuel mass

smooth-Figure 5 1T to 2T mode transitions with different valve

timing Two transitions are shown per plot: optimal (dashed)

and closing the WG 400 ms early (solid) Left: 3250 rpm

and 10 mg/stroke Right: 3250 and 30 mg/stroke Top: torque

evolution Bottom: valves’ position

Figure 6 2T to 1T mode transitions with different valvetiming Two transitions are shown per plot: optimal (dashed)and delaying the WG opening 200 ms (solid) See Figure 5caption for details

146 J GALINDO, H CLIMENT, C GUARDIOLA and J DOMÉNECH

depending on the operation mode For oscillations, several

possibilities are explored to improving the short-time torque

oscillations during the transition: adding corrections to the

instantaneous injected fuel mass, and using other strategies

in the VT2 opening profile The latter of these involves

slightly opening VT2 for transition preparation (it will be

referred as VT2 pre-lift) and slow actuation of VT2

As the fuel-to-air ratio is modified, opacity

measure-ments are provided in order to check the effect of the

strategy on the overall emissions

Finally, with the aim of covering different driving

condi-tions, transitions have been tested both in steady operation

(with fixed engine speed and injected fuel) and in engine

accelerations simulating road conditions

5.1 Injected Fuel Mass Corrections

The injected fuel mass correction strategy consists in

modi-fying the fuel injected during the transition Therefore, the

methodology of this strategy is to counteract the positive

torque oscillation peaks with negative pulses of fuel and

vice versa

The system is able to modify the nominal fuel by adding

3 fuel pulses that are completely independent and

configu-rable as illustrated in Figure 7 In addition, a steady

correc-tion value is added to compensate for the difference in

torque between 1T and 2T, which can be positive or

negative depending on the operating conditions

5.1.1 Engine steady conditions

Several amplitude and timing configurations for the

inject-ed fuel pulses were testinject-ed before achieving reasonable

performance In all cases, high accuracy was needed in the

correction timing, since injecting a positive fuel mass pulse

during a positive torque peak would result in an even more

problematic situation

Figure 8 illustrates the results of a transition with and

without the fuel correction strategy performed in steady

conditions, 2750 rpm and 200 Nm The transition was from

the 1T to 2T mode The top left graph shows that torque

evolution and an oscillation of 15 Nm (7.5%) is observed if

no fuel correction is applied The fuel correction evolution

during the transition is plotted in the bottom left graph The

fuel-to-air ratio and the exhaust gas opacity are shown inthe right-hand plots

In the previous case, since the fuel-to-air ratio is lowenough, the impact of this strategy is very small However,when the engine approaches full load, the situation becomescritical, since the engine operates near stochiometric condi-tions

Figure 9 illustrates a transition performed at full load,

2500 rpm and 365 Nm A variation of 40 Nm (11%) isobserved without the fuel correction strategy The fuelevolution that smoothes the engine torque is also repre-sented, where 15 mg/stroke additional fuel is necessary tocounterbalance the engine torque reduction However, thisstrategy has serious consequences at high load, where anopacity of 90% and a fuel-to-air ratio higher than 1.1appear because of the increase in fuel consumption.The 2T to 1T mode transitions are again depicted atpartial load (2750 rpm and 200 Nm) in Figure 10, and atfull load (2500 rpm and 365 Nm) in Figure 11 The transi-tion at partial load leads to an engine torque deterioration

of 15 Nm, which is similar to the value obtained for the

Figure 7 Fuel correction profile Labels indicate

configu-rable parameters for the initial fuel pulse

Figure 8 1T to 2T mode transitions at 2750 rpm and 200

Nm with (solid) and without (dashed) the fuel correctionstrategy Top left: torque Top right: fuel-to-air relative ratio.Bottom left: injected fuel Bottom right: opacity

Figure 9 1T to 2T mode transitions at 2500 rpm and fullload with (solid) and without (dashed) the fuel correctionstrategy See Figure 8 caption for details

STRATEGIES FOR IMPROVING THE MODE TRANSITION IN A SEQUENTIAL PARALLEL TURBOCHARGED 147

opposite mode transition Moreover, this backwards

transi-tion at full load results in a torque drop of 20 Nm, which is

comparably lower than what was observed for the 1T to 2T

mode transition

Since these transitions do not present strong oscillations

in the engine torque, the required fuel corrections in order

to smooth the engine operation achieve maximum values

of 3 and 5 mg/stroke at partial and full loads, respectively

Therefore, when using the correction fuel strategy, the

fuel-to-air ratio and exhaust gas opacity give slightly higher,

though still acceptable, values

5.1.2 Engine acceleration conditions

A transition during engine acceleration is quite different

from the steady operation equivalent: since the WG is

closed during the acceleration process, the engine torque

evolution only slightly increases when VT2 is activated

due to the reduction of the exhaust manifold pressure and

hence the pumping losses Although the torque oscillations

in these conditions are not as relevant as in steady

transi-tions, the fuel correction strategy has been also tested in

order to completely eliminate the torque peak

Figure 12 illustrates a transition under an engine ration The upper graph shows the torque curves with andwithout the fuel correction strategy The graph on the rightshows in detail the torque peak generated during the transi-tion, where 10 Nm of torque oscillation (4%) is detectedwhen no fuel correction strategy is applied Since the goal

accele-is to reduce the engine torque, only reductions in theinjected fuel pulses are applied, as depicted in the left-handbottom plot This prevents negative consequences fromarising due to increases in exhaust gas opacity

5.2 VT2 Pre-lift ActuationThe strategy is intended to improve 1T to 2T transition bymeans of an initial acceleration of TC2: in a first step, VT2does not open completely, but just enough to allow thepassage of the exhaust gases through T2 The purpose ofthis action is to accelerate TC2 and pressurize the ductsdownstream of C2 and upstream of VC2 Once TC2 islaunched, the second step is to completely open VT2 TheVT2 pre-lift actuation was tested as a fuel correction strategy

at both steady and transient operation

5.2.1 Engine steady conditionsFigure 13 right-hand plot replicates the conditions ofFigure 8 (2750 rpm and 200 Nm), and the left-hand plotthose of Figure 9 (2500 rpm and 365 Nm, full load) Byusing VT2 pre-lift in steady conditions, engine torqueoscillations are reduced, although slight fluctuations stillappear They correspond precisely to the two-step openingprofile of VT2, as shown in the bottom plots

At partial load, the maximum oscillation is reduced from

16 Nm at normal VT2 actuation, to 8 Nm (from 8% to 4%)

At full load the reduction is significant, from 40 Nm to 10

Nm (or from 11% to 3%) The duration of the VT2 pre-liftperiod must increase with engine load to properly accele-rate TC2

Figure 10 2T to 1T mode transitions at 2750 rpm and 200

Nm with (solid) and without (dashed) the fuel correction

strategy See Figure 8 caption for details

Figure 11 2T to 1T mode transitions at 2500 rpm and full

load with (solid) and without (dashed) the fuel correction

strategy See Figure 8 caption for details

Figure 12 1T to 2T mode transition during an engineacceleration Results are shown with (solid) and without(dashed) the fuel correction strategy Top left: torque Topright: detail of torque evolution during the transition Bottomleft: injected fuel Bottom right: opacity during the transition

148 J GALINDO, H CLIMENT, C GUARDIOLA and J DOMÉNECH

5.2.2 Engine acceleration conditions

As previously noted, engine torque oscillations in full-load

accelerations are not as important as they are in steady

operation Nevertheless, some improvement is possible with

pre-lift actuation Figure14 replicates the conditions of the

results of Figure 12 Engine torque differences at the early

stages of the acceleration process are due to slight

modifi-cations in the smoke limiting strategy between both tests

The figure shows the VT2 strategy in the left-bottom

plot, which completely smoothes torque evolution during

the transition

5.3 VT2 Slow Actuation

The last strategy evaluated to reduce the torque oscillations

during the transition consists in modifying the VT2

actua-tion speed

For 1T to 2T transitions, the main purpose of this

strategy is to avoid the sudden aperture of VT2, which

causes an abrupt drop of exhaust manifold pressure and

increases engine torque

A restriction in the vacuum system of the pneumaticactuator was used to modify the valve response time Thus,the effective diameter of the duct that links the actuatorchamber of the VT2 valve with the rest of the vacuumcontrol circuit was reduced, increasing the response time.Results for both steady and transient engine conditionsare shown

5.3.1 Steady engine conditionsThe left-hand plots in Figure 15 illustrate the transitionfrom 1T to 2T mode at high engine load, 2500 rpm and 300

Nm The standard VT2 aperture takes about 200 ms, whilethe duration exceeded 1 s using the slow VT2 actuation.The top plot shows a reduction in the amplitude of thetorque oscillation from 30 Nm to 20 Nm (from 10% to6.7%), whereas the duration of the oscillation is doubled.The transition from 2T to 1T mode at 2500 rpm and 300

Nm is represented in the right-hand plots of Figure 15 Inthis case, no significant improvements are obtained Sincethe VT2 is set in series with T2, VT2 effective area hasimportant effects on the flow when it is comparativelylower than the effective area of the turbine For the parti-cular design considered, this only occurs when the valve isnearly closed This justifies the effect on the 1T to 2T modetransition, since the valve opens slowly at the beginning ofthe transition and has little influence on the 2T to 1T modetransition, where it closes at the later stages

5.3.2 Engine acceleration conditionsThe VT2 slow actuation was also examined in engineacceleration conditions and the results are illustrated in theFigure 16

The results show that the small torque peak is

eliminat-ed However, the exhaust gas energy bypassed throughVT2 affects the boost pressure and also the combustionefficiency, yielding a loss in torque during the transitionprocess Thus, this strategy is not suitable for air-deficientfull-load accelerations

Figure 13 Effect of VT2 pre-lift actuation strategy (solid)

during a 1T to 2T transition at 2500 rpm and 365 Nm (left)

and at 2750 and 200 Nm (right) Dashed lines show

nominal transitions Top: torque Bottom: VT2 position

Figure 14 1T to 2T mode transition during an engine

acceleration Results are shown using the VT2 pre-lift

strategy (solid) and using nominal valve actuation (dashed)

See Figure 12 caption for details

Figure 15 Effect of the VT2 slow actuation strategy (solid)during a 1T to 2T (left), and a 2T to 1T (right) transition.Dashed lines show nominal transitions Top: torque Bottom:VT2 position

STRATEGIES FOR IMPROVING THE MODE TRANSITION IN A SEQUENTIAL PARALLEL TURBOCHARGED 149

6 CONCLUSIONS

Transition phenomena and optimized valve management of

a sequential parallel turbocharged engine have been shown

In addition, different strategies for improving the mode

transition were also presented and experimentally

com-pared The main conclusions of this comparative study are:

(1) A fuel correction strategy allows the complete

elimin-ation of torque oscillelimin-ations during the transition for

both steady and transient cases However high levels of

exhaust gas opacity are obtained when approaching full

load conditions, thus limiting the application range of

this strategy The fuel correction strategy also

compen-sates for the steady state torque difference between the

two operation modes

(2) The VT2 pre-lift strategy smoothes the torque

oscilla-tion in both steady and transioscilla-tion engine condioscilla-tions

without significant drawbacks However, small

oscilla-tions are still present

(3) The VT2 slow actuation strategy shows good results in

steady conditions, however it is not recommendable

under transient engine conditions

These strategies for smoothing the mode transition can

be combined In the final version of the engine, the VT2

pre-lift strategy was combined with the fuel correction

strategy to provide smooth transitions in both steady and

transient conditions (Galindo et al., 2009)

REFERENCES

Arnold, S., Groskreutz, M., Shahed, S M and Slupski, K

(2002) Advanced variable geometry turbocharger for

Benajes, J., Luján, J M., Bermúdez, V and Serrano, J R

(2002) Modelling of turbocharged diesel engines in

transient operation Part 1: Insight into the relevant

Automobile Engineering, 216,431–441

Benvenuto, G and Campora, U (2002) Dynamic tion of a high-performance sequentially turbochargedmarine diesel engine Int J Engine Research 3, 3, 115−

Galindo, J., Climent, H., Guardiola, C., Tiseira, A andPortalier, J (2009) Assessment of a sequentially turbo-charged diesel engine on real-life driving cycles Int J Vehicle Design, 49

Kench, J M and Klotz, H (2002) Model-based sequentialturbocharging optimization for series 8000 M70/M90

Luján, J M., Climent, H., Guardiola, C and García-Ortiz,

J V (2007) A comparison of different algorithms forboost pressure control in a heavy-duty turbocharged

Auto-mobile Engineering, 221,629–640

Payri, F., Benajes, J., Galindo, J and Serrano, J R (2002).Modelling of turbocharged diesel engines in transientoperation Part 2: Wave action models for calculating thetransient operation in a high speed direct injection engine

Prc Instn Mech Engrs Part D, J Automobile neering, 216, 479–493

Engi-Portalier, J., Blanc, J C., Garnier, F., Hoffmann, N., Schorn,N., Kindl, H., Galindo, J., Jeckel, D., Uhl, P and Laissus,J.-J (2006) Twin turbo boosting system design for thenew generation of PSA 2,2 liter HDI diesel engines

Proc Thiesel Conf 2006, 589−607

Ren, Z., Campbell, T and Yang, J (1998) Theoretical andexperimental study on the performance of a sequentially

and Air Management Systems, C554/010/98

Serrano, J R., Arnau, F J., Dolz, V., Tiseira, A., Lejeune,

M and Auffret, N (2008) Analysis of the capabilities of

a two-stage turbocharging system to fulfil the us2007anti-pollution directive for heavy duty diesel engines

Int J Automotive Technology 9, 3, 277–288

Tashima, S., Tadokoro, T., Okimoto, H and Niwa, Y (1991).Development of sequential twin turbo system for rotary

Tashima, S., Okimoto, H., Fujimoto, Y and Nakao, M.(1998) Sequential twin turbocharged rotary engine of

Internal Combustion Engine MacMillan Publishers Ltd Houndmills

Figure 16 1T to 2T mode transition during an engine

acceleration Results are shown using VT2 slow actuation

strategy (solid) and with nominal valve actuation (dashed)

See Figure 12 caption for details

International Journal of Automotive Technology , Vol 10, No 2, pp 151 − 160 (2009)

151

EXPERIMENTAL STUDIES ON LOW PRESSURE SEMI-DIRECT FUEL INJECTION IN A TWO STROKE SPARK IGNITION ENGINE

M LOGANATHAN 1) and A RAMESH 2)*

Indian Institute of Technology Madras, Chennai 600036, India

(Received 13 August 2007; Revised 6 October 2008) ABSTRACT− In this work a two-stroke scooter engine was modified to work with semi-direct injection of gasoline at a pressure of 8 bar from an injector in the cylinder barrel pointed toward the cylinder head The influence of injection timing, injection pressure, spark plug location and air-fuel ratio, on performance, emissions and combustion characteristics has been investigated In addition, a comparison has been made with manifold injection of gasoline on the same engine at a given speed and various outputs A significant reduction in HC emissions and fuel consumption with no adverse effects on NOx emissions and combustion stability was observed A small drop in power and increase in CO emission were observed disadvantages of the new injection system Injection timing was found to be the most important factor and a balance between reduction in short- circuited fuel by late injection, and time for mixture preparation by advancing the injection, was found to be essential.

KEY WORDS : Two stroke engines, Semi-direct injection, Emission control, Fuel Injection

1 INTRODUCTION

The demand for small capacity engines with high power to

weight ratio and low emissions is well known Though the

four-stroke engine has replaced its two-stroke counterpart

in many applications, two-stroke engines have been the

primary choice for vehicles like mopeds, small scooters,

snowmobiles and hand held power tools However, these

engines are also known for their high emissions of HC and

CO and fuel consumption These defects are due to

short-circuiting of the fuel supply during the scavenging phase

and to the dilution of the fresh charge by residual exhaust

gases The trapped residual charge is used to lower NOx

emissions If the two-stroke design can be improved to

reduce the short-circuiting of fuel its advantages may make

it more attractive for use in applications in which

four-stroke engines are the traditional choice Despite the

defi-ciencies inherent to current two-stroke designs, there are

still several two-stroke engine designs being manufactured

and used every year in applications such as those

mention-ed above There are also many old engines still in service

Hence, it will be desirable to devise methods that improve

the performance of current designs by simple modifications

and components that may allow retrofitting of existing

systems

Injection of fuel into the manifold of four-stroke engines

has become a highly-perfected technique that (in

conjunc-tion with a three-way catalytic converter) performs adequateemission control Such a technique will not yield signifi-cant results in a two-stroke engine, however, as the mainproblem of short-circuiting of the fuel will still be present.Injection of the fuel directly into the combustion chamber

is a desirable modification as it presents the possibility of

(1972) developed a mechanical fuel injection system for a

350 cc two-stroke SI (Spark Ignition) engine wherein theinjector could be located in different places, for example,the cylinder bore, head and transfer port (the port leadingthe fresh charge from the crank case to the cylinder)sequentially In this work, injection at the cylinder headwas found to give the best results An air assisted injectionsystem was developed for small outboard engines by

com-pressor and a combination of air and fuel injectors mounted

on the cylinder head The resultant highly-stratified fuel airmixture reduced the specific fuel consumption by 40% andexhaust emissions by 60%

Pierre and Stephane (1996) developed an air assistedfuel injection device called the IAPAC (Injection assisteeAir Comprime) system Here, the crankcase compressedair is used to push fuel that has been pre-injected into acavity into the engine The engine uses a mechanicallyoperated valve located on the cylinder head to inject the

(1991) have also developed air-forced injection systems

*Corresponding author. e-mail: aramesh@iitm.ac.in

152 M LOGANATHAN and A RAMESH

injection system that uses crank case compressed air and a

rotary injection valve Cobb (2001) developed an injection

system for a small two-stroke engine that injects a mixture

of air and fuel using the energy of a compression wave

that the injection spray type, spark plug location, injection

timing and fuel–air mixing are all very influential A wide

spray produces a well-mixed cloud of fuel in the vicinity of

the spark plug to improve combustion A narrow spray

produces a stratified air and fuel mixture near the spark

plug, resulting in unpredictable combustion Johnson and

Wong (1998) have performed a related experimental

investi-gation using a 1100cc 3 cylinder two stroke SI engine The

fuel injectors used in the system are accumulator type fuel

injectors with high-speed solenoid valves The engine is

capable of operating with very short injection durations

(less than 800 microseconds) with delivery a very low

amount of fue (22 mg) The fuel injector was located in the

cylinder head and the injection pressure was 140 bar The

performance of the engine revealed that the HC emissions

decreased by 76% while the fuel consumption decreased

by 42%

For two-stroke engines that must be run at high speeds,

achieving stable operation with direct injection is known to

be a difficult task Stability is typically accomplished using

costly equipment, either to inject fuel at high pressures, or

to provide an air assisted injection system Such

techno-logies have proven difficult to control and so expensive

that the cost advantage of the two-stroke engine design is

lost It is thus desirable to use injection equipment typically

used on four stroke SI engines to reduce the

short-circuit-ing of fuel in two stroke engines (thereby capitalizshort-circuit-ing on

the low cost of widely used technology) These injectors,

however, cannot be used for direct injection as they cannot

withstand high temperatures and pressures The pump used

also is not capable of developing the pressures necessary

for direct injection

Semi-direct injection has already been investigated as a

means to reduce short circuiting by several researchers

Douglas and Blair (1982)reported that injection of fuel in a

two-stroke engine at a pressure of 0.28 MPa from the

cylinder bore can lead to about a 30% reduction in fuel

consumption and a 60% reduction in exhaust emissions

Vieilldent (1978) investigated a cylinder bore injection

system and indicated that it could reduce emissions and

specific fuel consumption while manifold injection resulted

in performance figures similar to that of a carbureted system

Grasas-alsina et al (1986) investigated the performance

of low pressure fuel injection (0.3 MPa) at the transfer port

of a two stroke engine and concluded that that the fuel

consumption was reduced by 10% to 30% compared to a

carbureted engine The injection timing for lowest fuel

(2002) investigated cylinder wall injection in two-stroke

engines for using a 430 cc, single cylinder engine Theyfound that an injector position as close to TDC as possible

on the rear side of the transfer port is suitable and that theentire amount of fuel should be injected toward the top ofthe piston surface The optimal timing was found to be

pressure of 0.6 MPa A reduction in HC emission rangingfrom 40% to 90% and of fuel consumption by 15% to 35%was obtained

3 EXPERIMENTAL SETUP

A single cylinder two-stroke spark ignition scooter enginewhose specifications are shown in Table 1 was connected

to an eddy current dynamometer Fuel flow was measured

on a mass basis Air flow was measured by a turbine flowmeter Atmospheric pressure and temperature were used toconvert the volume flow rate of air to a mass flow rate Theair-fuel ratio that is reported is the overall value asmeasured at the inlet of the engine A lubricating oil supplysystem consisting of a plunger pump driven by a variablespeed motor was used

The lubricating oil was injected directly into the airstream leading to the crankcase Since it was found that, atcertain operating conditions, the fuel touched some portion

of the cylinder wall, a small amount of lubricant was alsoadded to the fuel to avoid piston seizure A potentiometerwas used to determine the throttle position

An infrared gas analyzer (HORIBA, MEXA 554J) wasused to measure the exhaust hydrocarbon (n-hexane equi-valent), carbon monoxide and carbon dioxide concen-trations The concentration of the NO in the exhaust wasdetected using a chemiluminescence analyzer A flushmounted piezoelectric pressure transducer was used to

EXPERIMENTAL STUDIES ON LOW PRESSURE SEMI-DIRECT FUEL INJECTION 153

obtain cylinder pressure signals The signals were captured

on a high-speed data acquisition system and then analysed

for computing heat release rate Heat release rate was

calculated from the cylinder gas pressure history (Lanzafame

and Messina, 2003) The schematic of the test setup is

shown in Figure1 An electronic circuit was used to vary

set the optimal value

The semi-direct fuel injection system is comprised of a

gasoline injector, a fuel pump, a pressure regulator and

associated electronic circuits for controlling the injection

timing and duration A positive-displacement pump was

used to supply gasoline at a pressure of 5 to 8 bar The

injector used was of the solenoid operated single hole typewith a pintle that gave a conical spray angle of 27o A holewas drilled in the cylinder barrel at an angle of 45o to thecylinder axis, opposite the exhaust port, to locate the fuelinjector This angle was selected so that the fuel wouldreach the spark plug without interfering with the cylinderwall

A photograph of the injector and fuel spray is seen inFigure 2 The injector holder was welded to the barrel andcare was taken to see that there was no deformation ofcylindrical barrel surface The level of the injector abovethe level of the exhaust port had to be carefully decided Ifthe injector is not well above the level of the exhaust portthere will be little flexibility in the injection timing On theother hand, if the hole is much above the exhaust port, theinjector will be exposed to high pressure and high temper-ature exhaust gases during the expansion stroke A benchtest on the injector used in this work indicated that it couldtolerate a back pressure of about 8 bar without any leak.Based on the cylinder pressure crank angle curve obtain-

ed during normal engine operation, the level of the holewas finally decided so that it would be completely closed at

multivib-Table 1 Specifications of the engine

Figure 1 Experimental setup

1 Engine, 2 Dynamometer, 3 Dynamometer controller, 4

Data acquisition system, 5 Pressure transducer, 6 DC

sensor, 7 Ice path, 8 Exhaust gas analyzer, 9 Engine

speed indicator, 10 Temperature indicator, 11 Fuel

injector, 12 Pressure regulator, 13 Pressure gauge, 14

Fuel pump, 15 Weighing machine, 16 Fuel tank, 17

Throttle body, 18 Throttle position sensor, 19 Surge tank,

20 Air flow meter, 21 Electronic control system, 22 12V

Battery, 23 Lubrication oil tank, 24 Lubrication oil pump,

25 Motor, 26 Air compressor

Figure 2 Photographic view of the injector hole and thespray (seen from the cylinder head side)

154 M LOGANATHAN and A RAMESH

rator-based electronic circuit was used to generate pulses at

the correct timing and for the correct duration to energize

the injector The output of the crankshaft position sensor

was conditioned and used to trigger this circuit The range

ranged between 1.5 ms and 6 ms The amount of fuel

injected for a given injection duration depends on the

supply pressure of the fuel and on the downstream pressure

of the injector The downstream pressure of the injector

rises if the injection extends into the region where the

exhaust port is closed

4 RESULTS AND DISCUSSION

The engine was first tested with the semi-direct injection

system at an injection pressure of 6 bar (0.6 Mpa) The

injection pressure was continuously monitored and

main-tained The effect of injection timing was initially

deter-mined at a representative throttle setting of 25% at a

con-stant speed of 3000 rpm Injection timings in the range of

addition, the injector opening time was found out to be

1 ms In all of the figures the start of injection actually

refers to the start of the pulse given to the injector, which

rpm to open The results obtained with the semi-direct

injection system have been compared with the results

obtained with a manifold injection system under similar

operating conditions The manifold injection results were

obtained on the same engine and have been discussed in an

4.1 Effect of Injection Timing, Injection Pressure and Spark

Plug Position

Changing the width of the pulse applied to the injector

varies the air-fuel ratio As mentioned earlier the speed was

held constant at 3000 rpm and the throttle was kept at 25%

The effects of power output and brake thermal efficiency

are seen in Figures 3 and 4 Very early injection will lead to

greater short-circuiting losses Late injection may result in

insufficient time for good mixture preparation We find

that, in the case of semi-direct injection, the injection

timings lead to lower power and thermal efficiency This is

probably predominantly because ocaused by having

insuffi-cient time to produce a combustible mixture The injection

duration, with respect to the amount of fuel injected per

cycle, is seen in Figure 5 with data for various injection

timings The curves within the dotted line indicate the

injection duration in milliseconds (ms) while while those

outside give the injection duration in crank angle degrees

bBDC, the injection duration extends from about 45o to 55o

crank angle In the case of the retarded injection timings the

thermal efficiency and the need to inject large quantities offuel In all cases the injection stops before both the transferand exhaust ports close Further, retardation of the injectiontiming leads to irregular operation as the injection occurs,even after the injection port closes As the injection timing

is advanced, peak power is generally shifted to leaner

bBDC were not tried because the exhaust port would only

results in a good power probably due to the formation of a

bBDC when the air-fuel ratio reduces, it leads to thehighest brake thermal efficiency on account of a reduction

in short-circuiting It also leads to higher power with lean

Figure 3 Variation of brake power with injection timing

Figure 4 Variation of brake thermal efficiency with tion timing

injec-EXPERIMENTAL STUDIES ON LOW PRESSURE SEMI-DIRECT FUEL INJECTION 155

26.8% at an air air-fuel ratio of 20.5:1, while the manifold

injection system gives a 25% efficiency at an air-fuel ratio

of 17:1 (Figure 4) Thus there is a significant improvement

in the thermal efficiency as compared to the manifold

injection method, which is mainly due to reduced

short-circuiting losses Since the short-circuited charge is mostly

air, the trapped air-fuel ratios in the case of semi-direct

injection are expected to be lower than the overall air-fuel

ratios indicated in the graphs The air-fuel ratios indicated

in the graphs are those obtained from the overall air and

fuel flow rates

Emissions of HC and CO are seen in Figures 6 and 7

There is a significant decrease in the HC level with

semi-direct injection due to the reduction in short circuiting In a

two-stroke engine, HC emissions will depend on how much

fuel is short-circuited and also on the extent of combustion

of the trapped fuel Hence, in the case of retarded injection

timings (even though a better trapping of the injected fuel

can be expected) the HC levels are higher due to poor

combustion because of insufficient time available for ture formation

minimum HC with manifold injection is at 1544 ppm with

an air-fuel ratio of 16.8:1 whereas it is only 600 ppm at anair-fuel ratio of 19:1 with semi-direct injection In the case

short-crcuiting, even though the differences in peak thermalefficiencies are not significant and power output is higher For any given air-fuel ratio, the CO level with semi-directinjection is higher than with manifold injection because thetrapped air and fuel mixture is richer With lean mixtures,where almost all of the trapped fuel can be expected toburn, we see that the CO level increases as we retard theinjection timing ie., as more fuel is trapped and the trappedmixture becomes richer (Figure 7)

The NO levels seen in Figure 8 are lowest for theFigure 5 Variation of fuel flow rate with Injection duration

Figure 6 Variation of HC emission with Injection duration

Figure 7 Variation of CO emission with Injection duration

Figure 8 Variation of NO emission with Injection duration

156 M LOGANATHAN and A RAMESH

manifold injection method where all of the air and fuel are

premixed and the charge temperature is lowest This leads

to attaining lower peak temperatures in this case

The peak also occurs at a slightly leaner mixture than

stoichiometric in this case With an injection timing of

level is also the highest With an injection timing of 49oCA

bBDC, NO levels are also significantly higher than those

from the manifold injection technique Also, in the case of

level occurs when the overall mixture is lean This is

because the trapped mixture is richer

The exhaust gas temperature, as seen in Figure 9, is

highest at any given air-fuel ratio for the case with the most

retarded injection timing Exhaust temperatures are lowest

under the manifold injection technique where the

short-circuiting of fuel will be the highest As we retard the

injection timing, the amount of short-circuited fuel will be

reduced and this is probably the reason for the increase in

the exhaust gas temperature (which is actually the

temper-ature of the exhaust, short-circuited air and fuel) The mass

of fuel injected per cycle with different injection timings is

shown in Figure 5

In all cases the mass of fuel injected at any air-fuel ratio

is higher with the semi-direct injection system This means

that the mass of air inducted is higher In the case of the

manifold injection system, this is because the vaporized

fuel displaces some of the air We find that the difference is

reduced as we move to leaner mixtures Based on the

thermal efficiency and HC emissions, an injection timing

release rates at different injection timings (at air-fuel ratios

that correspond to about an output of 2.3 kW) are

result in the highest heat release rates This also means that

combustion is completed in a very short time, leading to

good thermal efficiency Even though the heat release rate

is slightly higher with an injection timing of 72oCA bBDC the short-circuiting losses are expected to be lower with49oCA bBDC, leading to better efficiency The COV of

IMEP with different timings (seen in Figure 11) shows that

values indicate that the engine is quite stable at both ofthese injection timings

In subsequent experiments the injection pressure wasincreased to study the influence of fuel atomization Pre-ssures over 8 bar could not be tried due to the limitationsposed by the pump and injector These tests were also done

at 25% throttle, 3000 rpm and at an injection timing of

power, thermal efficiency and emissions is seen in Figures(12)~(15) In between 6 and 8 bar, there is very little effectseen in these parameters (there is a small decrease in the

HC and CO emissions) Hence, in all subsequent ments, the injection pressure was held at 8 bar Three sparkplug positions were tried (as shown in Figure 16) The airmovement in the cylinder will depend on the angle atwhich the fuel spray is injected and this will affect the final

experi-Figure 9 Variation of Exhaust gas temperature with

EXPERIMENTAL STUDIES ON LOW PRESSURE SEMI-DIRECT FUEL INJECTION 157

path of the fuel droplets It is necessary to ensure that there

is always a combustible mixture near the spark plug for

proper ignition As seen in Figures 17 to 19 the central

spark plug location leads to good thermal efficiency andlow emissions of HC and CO In fact, except for HCemissions, the spark plug location was not found to be a theparameter of much influence

4.2 Comparison of the Semi-direct Injection System withManifold Injection

Experiments similar to those conducted at 25% throttlewere conducted (for the semi-direct injection configu-ration) at throttle positions of 5%, 10%, 15%, 25%, 40%,50% and 100% In all of these tests the injection pressure

spark plug was located at the center of the cylinder head

At each throttle position the air-fuel ratio that resulted inthe best brake thermal efficiency was chosen as the optimumvalue The results obtained with the semi-direct injection

Figure 12 Variation of Brake power with injection pressure

Figure 13 Variation of Brake thermal efficiency with

injec-tion pressure

Figure 14 Variation of HC emission with injection pressure

Figure 15 Variation of CO emission with injection pressure

Figure 16 Different locations of the spark plug

158 M LOGANATHAN and A RAMESH

system at these optimal conditions have been comparedwith those of the manifold injection system in subsequentgraphs Detailed results for the manifold injection configu-ration on this same engine were reported in earlier work14

In short, the two methods have been compared at the bestair-fuel ratios and at every throttle position in subsequentgraphs

Figure 20 indicates that there is a significant ment in the brake thermal efficiency with the semi-directinjection system at all outputs (throttle positions) due to thereduction in the short-circuiting losses The highest brakethermal efficiencies are 23% and 28% for the manifoldinjection system and the semi-direct injection system,respectively The maximum improvement is expected atfull throttle where charge short-circuiting is the highest.However, this does not happen since the ports are openwhen injection occurs, even in the semi-direct injectionsystem Though the spray is not directed towards the ports,short-circuiting cannot be avoided in this case We alsoobserve that the maximum power is reduced in the semi-direct injection configuration This is because of the limitedtime available for fuel injection If higher power outputsare needed, the fuel has to be injected earlier and this willincrease the short-circuiting losses The maximum power is3.2 kW, compared to 3.4 kW attained with the manifoldinjection technique

improve-As seen earlier the CO levels are always higher with thesemi-direct system on account of the higher trapped air-fuel ratios We find a considerable reduction in the HClevel (as seen in Figure 21) at full throttle where short-circuiting losses dominate There is a reduction in HCemissions from 3000 to 728 ppm At very low throttle posi-tions, where HC is mainly due to incomplete combustion,the improvement is not significant Better combustion andhigher temperatures lead to higher NO levels in the case ofsemi-direct injection (Figure 22) However, the levels arenot alarming in comparison to those observed using the

Figure 17 Variation of Brake thermal efficiency with spark

EXPERIMENTAL STUDIES ON LOW PRESSURE SEMI-DIRECT FUEL INJECTION 159

manifold injection method The overall air-fuel ratios

(measured by the air and fuel flow rates) for best operation

are shown in Figure 23 Here we see that, as expected, the

overall mixture is leaner with the semi-direct injection

system (the trapped mixture is richer since the

short-circuited charge is mainly air) Wherever short-circuiting dominates, ie at high throttle settings, the difference issignificant At full throttle the best air-fuel ratio (as regards

Figure 21 Comparison of HC emission

Figure 23 Comparison of Air-fuel ratio

Figure 24 Comparison of peak cylinder pressure

Figure 25 Comparison of heat release rate

Figure 26 Comparison of COV of IMEP

160 M LOGANATHAN and A RAMESH

thermal efficiency) with the semi-direct injection is 23.6:1

as compared to 17.4:1 with the manifold injection system

There is no significant change in the peak cylinder

pre-ssure, as seen in Figure 24

The heat release rate curve indicates that combustion is

completed earlier in the cycle with the semi-direct injection

system (Figure 25) We also find that, at most throttle

sett-ings, the engine is more stable in operation with the

semi-direct injection system (as given by the COV of IMEP,

Figure 26)

5 CONCLUSIONS

Based on the tests conducted on the two stroke engine at

3000 rpm and at different outputs, the following

conclu-sions may be drawn:

There is a significant decrease in the HC level and

increase in the brake thermal efficiency with the

semi-direct injection as compared to manifold injection due to a

reduction in short-circuiting Combustion is also seen to be

stable

The peak brake thermal efficiency rises from 23% to

28% The HC level at full throttle drops from 3000 to 728

ppm and the NO levels are slightly increased

Injection timing is very critical in ensuring proper engine

operation in the semi-direct injection mode In the present

engine, the most suitable injection timing and injection

though retarded injection can lower short-circuiting losses,

it leads to poor thermal efficiency due to limited time

available for mixture formation The location of the spark

plug only affected the HC emissions level in this case

At any given equivalence ratio, the CO level with

semi-direct injection is higher than that with manifold injection

because the trapped air and fuel mixture is richer The peak

power output is also lower

At full throttle the best overall equivalence ratio

(mea-sured with inducted air and fuel) using the semi-direct

injection system is 0.65 as compared to 0.85 when using

the manifold injection system The delivery ratio of the

engine is higher with the semi-direct injection system at

any particular throttle setting

On the whole, this method is simple to implement and

can even be used for retrofitting small two stroke engines

for emission reduction

ACKNOWLEDGEMENT− The authors are thankful to The

Ministry of Human Resource Development, Government of India,

for the financial support for this work.

Paper No 850483

Gentili, R., Frigo, S and Tognotti, L (1994) Development

of a pumpless air assisted injection system for two-cycle

Grasas-Alsina, C., Freixa, E., Esteban, P and Masso, J.(1986) Low pressure discontinous gasoline injection in

Johnson, D E and Wong, H (1998) Electronic direct fuelinjection system applied to a 1100cc two-stroke person-

Lanzafame, R and Messina, M (2003) ICE gross heatrelease strongly influenced by specific heat ratio values

Int J Automotive Technology 4, 3, 125−133

Leighton, S., Cabis, M and Southern, M (1994) The OCPsmall engine fuel injection system for future two stroke

Loganathan, M., Manivannan, P V and Ramesh, A (2006).Investigations on performance and emissions of a twostroke SI engine fitted with a manifold injection system

Indian J Engineering and Materials Sciences, 13, 95−

102

Motoyama, Y., Takahiro, S., Akihiko, O and Takeo, Y.(2002) Emission reduction by cylinder wall injection in

Pierre, D and Stephane, V (1996) Automotive calibration

of the IAPAC fluid dynamically controlled two stroke

Schechter, M M and Levin, M B (1991) Air forced fuel

Paper No. 910664

Syvertsen, M L., Martin, J K., Hoffman, J A., Coats, S

W and Mc Ginnity, F A (1996) Injection and ignitioneffects on two stroke direct injection emissions and

Vieilledent, E (1978) Low pressure electronic fuel injection

Yamagishi, G., Tadanori, S and Hiroyoshi, I (1972) Astudy of two stroke cycle fuel injection engines for ex-

International Journal of Automotive Technology , Vol 10, No 2, pp 161 − 166 (2009)

161

EFFECTS OF VALVE TIMING AND INTAKE FLOW MOTION CONTROL

ON COMBUSTION AND TIME-RESOLVED HC & NOx FORMATION

CHARACTERISTICS

C L MYUNG 1) , K H CHOI 1) , I G HWANG 1) , K H LEE 2) and S PARK 1)*

(Received 30 April 2008; Revised 13 October 2008) ABSTRACT− In SI engines, valve events have a major influence on volumetric efficiency, fuel economy and exhaust emissions Moreover, swirl and tumble motions in the intake charge also improve combustion speed and quality by stratifying the mixture as well as intensifying the mixing rate of air and fuel This paper investigates the behaviors of an engine and the combustion phenomenon for various intake valve timings and intake charge motions using CVVT system and port masking schemes Test condition includes a part load and a cold idle condition inclusive of a cold start of the engine Time-resolved

HC and NOx emissions were also measured at an exhaust port to examine their formation mechanisms and behaviors with fast response HC/NOx analyzers In conclusion, the fast burning of fuel and improved combustion quality by enhanced charge motions reduced unburned HC emissions, and advancing the intake valve opening reduced HC as well as NOx Furthermore, HCs during the cold transient phase and idle conditions decreased with recalibrated start parameters such as lean air-fuel ratio and spark retardation via the enhancement of intake charge motions.

KEY WORDS : Port masking, Swirl, Tumble, CVVT, Fast response FID, Fast response CLD

1 INTRODUCTION

During the last several years, significant efforts have been

focused on the development of environmentally friendly

vehicles in the automotive industry to meet increasingly

stringent emissions legislation More than 95% of

hydro-carbon emissions during the FTP (Federal Test Procedure)

before catalyst activation in SULEVs (Super Ultra Low

Emission Vehicles) Therefore, a number of studies have

been carried out for reduction of cold start emissions

such as the use of a thin wall catalyst, flow optimized

exhaust manifolds and stainless steel exhaust manifolds

(Zhao, 2007)

CVVT (Continuously Variable Valve Timing) systems

are widely adopted since they can enhance engine

per-formance and simultaneously reduce exhaust emissions

and fuel consumption With the CVVT system, maximum

torque and power can be improved by optimizing valve

timing according to overall engine operating conditions

(Lee et al., 2007; Parvate-Patil et al., 2003) Intake valve

timing influences internal EGR (Exhaust Gas

Recircu-lation), which can reduce NOx emissions and fuel

con-sumption with longer valve overlap at part loads.Turbulence-intensified intake flow can improve combus-tion stability by reducing cycle-by-cycle variations As aresult, cold emissions can be greatly reduced by extendingthe LML (Lean Misfire Limit) and retarding the sparktimings during the cold start before catalyst activation

investi-gations on reinforced turbulence have been conducted,including distinct valve lifts, asymmetry ports and portthrottling (Roberts and Stanglmaier, 1999)

This study aims to examine the combustion phenomenafor various intake valve timings and intake charge motions

to achieve low emissions at the cold idle and transient coldstart phases of a gasoline engine To investigate the for-mation characteristics and reduction mechanisms of HCand NOx, high-resolution gas analyzers were used tomeasure time-resolved HC and NOx emissions

2 EXPERIMENTAL APPARATUS

A 1.5 L gasoline engine that had a CVVT system at theintake valve train was used in this study Table 1 shows thespecifications of the test engine

The port masking schemes shown in Figure 1 wereapplied at the intake port, substituting a conventional gas-ket to control the intake flow motions From baseline (a),fifty percent of the port area was screened to induce swirl

*Corresponding author. e-mail: spark@korea.ac.kr

162 C L MYUNG et al.

(b) and tumble effects (c), and 75% to induce a tumble

effect (d)

coolant temperature and a stoichiometric air/fuel ratio in

the part load experiment, and 1200 rpm/idle conditions and

Cylinder pressure was measured by a spark-plug-type

pre-ssure transducer installed at the 4th cylinder Crank

angle-resolved HC and NOx emission characteristics were

mea-sured at the exhaust port using FR-FID (Fast Response

Flame Ionization Detector) and FR-CLD (Fast Response

sampling point was 15 mm away from the exhaust valve at

the 4th exhaust port, which was separated from the other

three ports to prevent interference, especially for

3 RESULTS AND DISCUSSION

3.1 Part Load Experiment

Figure 2 shows the maximum cylinder pressures during

150 cycles vs the crank angle at which the maximum

pressures occurred Enhanced charge motions resulted in

higher and more concentrated peak pressures closer to the

fast burn line This means that the charge motion control

enhanced the flame propagation speed and combustion

stability by rapid burning of fuels

The flame development period is represented in Figure 3

according to the fraction of the fuel mass burned The

respectively The 75% tumble case resulted in 17.2o to burn

base condition Furthermore, initial flame development

was delayed with the 75% tumble due to the retarded spark

timing, but the combustion completed faster than the baseand swirl applied cases MBT (Minimum spark advance forBest Torque) was retarded with the application of the portmaskings due to the enhanced burning speed of the fuel.MBT values for the various types of post masking and theintake valve timing are given in Table 2

The combustion stabilities indicated by variations inIMEP (Indicated Mean Effective Pressure) over 50 cyclesare shown in Figure 4 for various intake valve timings andtypes of port masking As the intake open timings areadvanced, more burned gases flow back into the intake port

Table 1 Specifications of test engine

Figure 1 Port masking schemes

Figure 2 Maximum cylinder pressures of 150 cycles withrespect to the crank angle of Pmax (S/A=MBT, IVO=BTDC 10o, λ=1.0)

Figure 3 Mass burned fraction for various types of port

Table 2 MBT values for various port maskings and IVOtimings

EFFECTS OF VALVE TIMING AND INTAKE FLOW MOTION CONTROL ON COMBUSTION 163

and engine stability decreases As a result, the COV of

stabilities increased with enhanced charge motions, so the

COV was stabilized at less than 0.4% with the 75%

In conclusion, the intake charge motion control improved

combustion stability by rapid burn of fuels, which made it

possibile to reduce the engine-out HCs relative to the

conventional engine

Figure 5 indicates cyclic HC emissions with respect to

the port masking schemes measured at the exhaust port

with a fast response gas analyzer The engine-out HC

haust characteristics are described below When the

ex-haust valves open, blow down induces an increase in HC

concentration After the peak value, HC decreases by

re-oxidation with the excess oxygen in the port, and finally

increases once again just before the exhaust valves closebecause of the unburned HCs in the crevice volume andvalve seats The ‘real’ and ‘transit’ in the figure indicate thesampling delay of the gas analyzer, which was about 67°

by crank angle In comparison to the base engine, the 50%swirl-type masking resulted in a decrease of 12% in hydro-carbon, while the 50% tumble-type caused a 7% decreaseand the 75% tumble-type caused a 16% decrease on aver-age These reductions were caused by the reinforced intakecharge flow motions which increased flame propagationspeeds and enhanced combustion quality

Figure 6 represents cyclic NOx emission characteristicswith respect to the intake flow motions At the early stageafter the exhaust valve opens, blow down led the first peak

of the NOx concentration For an upward motion of thepiston, NOx showed the main peak value during the latterhalf of the exhaust stroke In contrast to the HCs, NOxincreased with the port masking, including both the swirland tumble types The enhancements in flame speed andcombustion quality made flame temperatures higher duringthe combustion period, so more nitrogen was oxidized As

a result, NOx concentration increased by 7%, 3% and 21%

on average with 50% swirl, 50% tumble, and 75% tumble,respectively

The crank angle-resolved hydrocarbon emissions ing to intake timings are shown in Figure 7 During thevalve overlap period, some of the unburned hydrocarbonsflowed back into the intake system, recirculated during asubsequent cycle, and went through the combustion pro-cess again As a result, the mean hydrocarbon emissionduring one cycle was reduced to 78% with an increased in

30o)

Backflow to the intake port during the valve overlapperiod also makes flame temperature lower So, NOxconcentration decreased from 1280 ppm to 700 ppm inaverage value for the most extended overlap period, asshown in Figure 8 In conclusion, increasing valve overlap

by advancing intake valve timing was an effective way to

Figure 4 IMEP variations with respect to intake valve

timing (upper, S/A=MBT, w/o port masking) and charge

Figure 5 Time-resolved HC emissions for various types of

Figure 6 Time-resolved NOx emissions for various types

164 C L MYUNG et al.

reduce engine-out raw NOx emission and HC emission

when the port masking was adopted

3.2 Cold Idle Experiment

Crank angle-resolved hydrocarbon and NOx emissions

with respect to the intake flow motions during the cold idle

condition are indicated in Figure 9 and 10, respectively In

particular, the target values of the excess air ratio and spark

advance were recalibrated with the port masking to

enhance cold emissions with stable engine operation The

excess air ratios were controlled to 0.9 in the base engine,

1.0 in the 50% swirl and tumble cases and 1.05 in the 75%

in the base engine case

With leaner A/F ratios and more retarded ignition

tim-ings, HCs were decreased by about 35% with 50% tumble,

50% with 50% swirl and 42% with 75% tumble On the

other hand, NOx increased about 27% with swirl and 43%

with the two tumble cases due to the leaner air/fuel

mix-ture Furthermore, ISA (Idle Speed Actuator) was opened

more to maintain the target engine speed due to thedecrease in the engine-out torque for the more retardedspark ignitions Therefore, more air was induced within thecylinder, which contributed to the NOx increase

3.3 Cold Start ExperimentFigures 11 to 15 indicate pressures, excess air ratio andemission characteristics during 4 seconds of the cold startstage For the base engine shown in Figure 11, the lowestvalue of the excess air ratio during the start phase was 0.72

the idle engine speed was stabilized Accumulated

were 39.6 kppmc and peak NOx was 145 ppm In Figure

12, the amount of fuel injection was controlled to be 85%

of the base engine level However, a leaner mixtureproportion without flow motion control was detrimental tothe HCs due to partial burn or even misfires As mentionedabove, enhanced charge motions on intake flow can extendthe lean misfire limit and the spark retardation limit Thus,

in the cases of 50% swirl and tumble motion (Figure 13and 14), the amount of fuel injection was reduced by 15%

Figure 7 Time-resolved HC emissions for various intake

masking)

Figure 8 Time-resolved NOx emissions for various intake

masking)

Figure 9 Time-resolved HC emissions for various types of

Figure 10 Time-resolved NOx emissions for various types

EFFECTS OF VALVE TIMING AND INTAKE FLOW MOTION CONTROL ON COMBUSTION 165

and an additional 10o of spark retardation was achieved As

a result, the accumulated HCs during the start stage were

36.6 kppmc with 50% swirl and 28.8 kppmc with 50%

tumble, which represented 7.6% and 27.3% reductions

with respect to the base engine The 75% tumble case

shown in Figure 15 reduced fuel injection by 25% with

more intensified intake flow motion Compared to the base

engine, the minimum excess air ratio was controlled to be

0.88 and the accumulated HCs during start decreased to

24.8 kppmc, a reduction of 37.4% Additionally, the

aver-age value of HCs at the stable idle staver-age (after 3 sec)decreased by 4% with 50% swirl, by 40% with 50% tumbleand by 50% with 75% tumble, compared to the baseengine’s 20 kppmc Otherwise, NOx values were increased

in all of the charge motion control cases due to the leanermixture proportions and improved combustion quality There-fore, when the enhanced flow motions were adopted duringstart, the increase in valve overlap would be considered toinhibit NOx emissions; nevertheless, conventional enginesgenerally use most retarded valve timings for startability.Cumulative HC within the start stage, peak NOx valuesduring start and mean exhaust values during stable idle aregiven in Table 3

Figure 11 Cylinder and intake pressure, excess air ratio,

HC and NOx during the cold start phase (base engine)

Figure 12 Cylinder and intake pressure, excess air ratio,

HC and NOx during the cold start phase (base engine with

lean mixture)

Figure 13 Cylinder and intake pressure, excess air ratio,

HC and NOx during the cold start phase (50% swirl)

Figure 14 Cylinder and intake pressure, excess air ratio,

HC and NOx during the cold start phase (50% tumble)

Figure 15 Cylinder and intake pressure, excess air ratio,

HC and NOx during the cold start phase (75% tumble).Table 3 HC and NOx value during cold start

HC (kppmc)

Start

NOx (ppm)

Peak

166 C L MYUNG et al.

4 CONCLUSIONS

The effects of intake valve timing and intake flow motion

control on combustion and exhaust emission characteristics

were investigated The results are summarized below

(1) The enhanced intake charge motions by the port

mask-ing scheme enhanced the flame propagation speed As

a result, the 10% to 90% fuel mass burned angle was

tumble-type port masking Accordingly, the port masking

stabilized combustion with rapid burning of fuels and

demonstrated the possibility to reduce engine-out

un-burned hydrocarbons

(2) Raw HC emission decreased by 12% with the 50%

swirl-type post masking, 7% with the 50% tumble-type

masking, and 16% with the 75% tumble-type masking

relative to emissions from the base engine in the part

load However, NOx increased up to 21% with 75%

tumble The extension of valve overlap with the

advan-ced intake openings was also an effective way to

reduce HC and NOx due to the re-burning of unburned

HC and the lowered flame temperature

(3) Hydrocarbon emissions decreased by a maximum of

50% due to a leaner mixture proportion and spark

retardation with enhanced charge motions during cold

idle operation On the contrary, NOx increased by 27%

with swirl and by 43% with tumbles due to the lean air/

fuel ratio and increase of air within the cylinder

Therefore, advance of the intake valve timing would be

required to prevent NOx emission with the application

of flow motion control

(4) During the cold start of the engine, HC was reduced by

a maximum of 37.4% at the start stage and by 50% at

the idle stage because of the lean mixture and the

additional spark retardation by port masking, which

were difficult to apply to the base engine On the other

hand, NOx increased by about 250% due to the

improved combustion and lean condition

Consequent-ly, intensified intake charge motions had a significant

effect on the reduction in HCs during the start of the

engine, although countermeasures to inhibit NOx

emi-ssion were needed

ACKNOWLEDGEMENT− This study was supported by the

BK21 and the Ministry of Knowledge Economy.

Kim, D S and Cho, Y S (2006) Idle performance of an

SI engine with variations in engine control parameters

Int J Automotive Technology 7, 7, 763−768

Kwak, H., Myung, C L and Park, S (2007) Experimentalinvestigation on the time resolved THC emission charac-teristics of liquid phase LPG injection (LPLi) engineduring cold start Fuel, 86, 1475−1482

Lee, H B., Kwon, H and Min, K (2007) Effects of ous VVA systems on the engine fuel economy andoptimization of a CVVT-VVL SI engine using 1Dsimulation Int J Automotive Technology 8, 6, 675−685.Myung, C L., Kwak, H., Hwang, I G and Park, S (2007).Theoretical flow analysis and experimental study ontime resolved THC formation with residual gas in a dualCVVT engine Int J Automotive Technology 8, 6, 697−

vari-704

Ohm, I Y and Park, C J (2006) Effect of fuel cation on initial flame development: Part 1-without swirl

stratifi-Int J Automotive Technology 7, 5, 519−526

Parvate-Patil, G B., Hong, H and Gordon, B (2003) Anassessment of intake and exhaust philosophies for vari-

Paper No 20034378

Peckham, M., Hands, T., Burrell, J., Collings, N and Schurov,

S (1998) Real time in-cylinder and exhaust NO

980400

Roberts, C E and Stanglmaier, R H (1999) Investigation

of intake timing effects on the cold start behavior of a

Santoso, H and Cheng, W K (2002) Mixture preparationand hydrocarbon emissions behaviors in the first cycle

Sung, J M., Kim, H W and Lee, K H (2002) Analysis

on cycle-by-cycle NO emissions from an SI engine with

Engineers 10, 2, 73−79

Gasoline-Powered Vehicles SAE Handbook

International Journal of Automotive Technology , Vol 10, No 2, pp 167 − 174 (2009)

167

DESIGN OF ANFIS NETWORKS USING HYBRID GENETIC AND SVD METHODS FOR MODELING AND PREDICTION OF RUBBER ENGINE

MOUNT STIFFNESS

J MARZBANRAD * and A JAMALI

Automotive Engineering Department, Iran University of Science & Technology, Narmak 16846-13114, Tehran, Iran

(Received 9 August 2007; Revised 19 July 2008)

ABSTRACT− Genetic Algorithm (GA) and Singular Value Decomposition (SVD) are deployed for optimal design of both the Gaussian membership functions of antecedents and the vector of linear coefficients of consequents, respectively, of ANFIS networks These networks are used for stiffness modelling and prediction of rubber engine mounts The aim of such modelling

is to show how the stiffness of an engine mount changes with variations in geometric parameters It is demonstrated that SVD can be optimally used to find the vector of linear coefficients of conclusion parts using ANFIS (Adaptive Neuro-Fuzzy Inference Systems) models In addition, the Gaussian membership functions in premise parts can be determined using a GA.

In this study, the stiffness training data of 36 different bush type engine mounts were obtained using the finite element analysis (FEA).

KEY WORDS : ANFIS (Adaptive Neuro-Fuzzy Inference Systems), Engine mount; Genetic algorithms (GAs), SVD (Singular Value Decomposition), FEA (Finite Element Analysis)

1 INTRODUCTION

Process modelling and system identification using

input-output data have always been the focus of many research

efforts Astrom and Eykhoff (1971) described system

iden-tification techniques that are applied in many fields to

model and predict the behaviours of unknown and/or very

complex systems based on given input-output data

Theoreti-cally, to model a system, a precise understanding of the

explicit mathematical input-output relationship is required

Alternatively, Sanchez et al (1997) presented

soft-comput-ing methods that involve computation in imprecise

environ-ments and have gained significant attention The main

components of soft computing, namely, fuzzy-logic, neural

networks, and genetic algorithms, have a great ability to

solve complex non-linear system identification and control

problems Among these methodologies, Porter and

Nariman-Zadeh (1994) proposed evolutionary methods that have

mostly been used as effective tools for both system

identifi-cation and optimal design of fuzzy and neural network

systems Fuzzy rule-based systems have been an active

field of research because of their unique ability to build

models based on experimental data Lee (1990) presented

the concept of fuzzy sets, which deal with uncertain or

vague information, and paved the way for applying them to

real and complex tasks Indeed, fuzzy-logic, together with

rule-based systems, has the ability to model the approximate

and imprecise reasoning processes, which are common inhuman thinking or human problem solving This results in

a policy that can be mathematically evaluated using fuzzyset theory Therefore, Wang (1992) stated that fuzzy systemscould be effectively employed as universal approximators

to perform input-output mapping Porter and Zadeh (1995) showed that such fuzzy systems could beiteratively designed using different evolutionary searchmethods, where such genetic-fuzzy systems continue togrow in visibility, as noted by Cordon et al (2001) In fact,these fuzzy systems are trained by examples (X i, y i) (i=1, 2,

Nariman-…, m) in terms of input-output pairs Recently, Wang et al.(1999) proposed the use of a combination of orthogonaltransformation and back propagation methods to train acandidate fuzzy model and to remove its unnecessary fuzzyrules In other recent works, Darvizeh et al. (2001) showedthat Singular Value Decomposition (SVD) can be used toenhance the performance of both fuzzy and GMDH-type(Group Method of Data Handling) neural network modelsobtained using simple heuristic approaches In such networks,every two input neurons are connected to produce a hidden

or output neuron using a linear or, more commonly, a linear quadratic form of function

SVD in combination with a genetic algorithm to optimallydesign a fuzzy system for modelling purposes that demon-strated its superior performance in comparison with previ-ous works However, a fuzzy model consisting of a largenumber of IF-THEN rules to map inputs to outputs is not

*Corresponding author. e-mail: marzban@iust.ac.ir

168 J MARZBANRAD and A JAMALI

desired due to the phenomenon of over fitting, which

reduces the generalising property of the fuzzy model to

predict the unforeseen data Similarly, the

Takagi-Sugeno-Kang (TSK) type fuzzy models are widely used for control

and modelling because of their high accuracy and relatively

small model size In the TSK models, which are also known

as neuro-fuzzy systems, the consequents of the fuzzy rules

are explicit functions, usually with linear relationships, of

the input variables rather than fuzzy sets In other words, as

presented by Hoffmann and Nelles (2001) the crisp linear

relation portion of the consequents of a TSK fuzzy rule

describes the underlying model in the local

multi-dimen-sional region specified in the premise part of that fuzzy

rule Therefore, two types of tuning procedures are required

for proper partitioning of the input space and number of

fuzzy rules, known as structural tuning, and for the

para-meters of the consequent parts of the fuzzy rules, known as

parametric tuning In recent years, different approaches

have been adopted for optimal tuning of such models based

on either heuristic search or fuzzy clustering for the

pre-mise part and least squares for linear parameters in the

conclusion of the fuzzy rules Genetic algorithms have

received a great deal of attention for the optimal selection

of the premise part of TSK type fuzzy rules in Wang and

Yen’s 1999 literature In order to identify the parameters of

the consequents, there have been attempts in the literature

to use SVD as a linear optimisation technique Jang (1993)

proposed an equivalent approach to the TSK models as an

Adaptive Neuro-Fuzzy Inference System, ANFIS In this

model a hybrid learning method is used for tuning

para-meters in both antecedents and consequents of embodied

TSK-type fuzzy rules

The engine is the largest concentrated mass in a vehicle

and will cause vibration in the vehicle’s body if it is not

properly isolated and constrained An ideal engine mount

system isolates engine vibration caused by unbalanced

forces and prevents engine bounce from shock excitation

Yu et al (2001) successfully used modern engine mounting

systems to isolate the driver and passenger from both noise

and vibrations generated from the engine The proper design

of rubber mounts may be the most effective engineering

approach to improving the ride of a vehicle The analysis of

engine mounting rubber components should be

accompani-ed by an analysis of the vibration of the engine mount

system Bernuchon (1984) showed that it is necessary not

only to know the properties of the rubber and where to

place the mounts but also to determine the optimum design

of a rubber part to achieve the desired properties along with

the required load-bearing capacity resulting from the

system vibration analysis

This paper models the stiffness variation with the

geo-metric parameters of a typical engine mount using the

ANFIS network Using this method, 36 different geometry

bush type engine mounts are analysed by FEA to obtain

their corresponding stiffness in 3orthogonal directions (k x,

engine mount are regarded as inputs, whilst the stiffnessvalues computed by FEA are regarded as outputs TheANFIS network identifies the input-output relationship that

is a set of TSK-type fuzzy rules for the modelling of theengine mount stiffness Such an ANFIS identification pro-cess needs, in turn, optimisation methods to find both theGaussian membership functions of the antecedents and thevector of linear coefficients of the consequents For thisreason, a hybrid genetic algorithm and SVD are used forthe optimal selection of Gaussian membership functions ofpremise parts and linear parameters of the ANFIS’s conclu-sion part, respectively

2 MODELLING USING ANFIS

An ANFIS consisting of a set of TSK-type fuzzy IF-THENrules can be used in modelling to map inputs to outputs.The formal definition of such identification problem is tofind a function so that it can be approximately usedinstead of the actual one, f This is done in order to predictoutput for a given input vector X=(x 1, x 2, x 3, , x n) asclose as possible to its actual output y Therefore, given m

observations of multi-input-single-output data pairs so that

(3)

In this way, a set of linguistic TSK-type fuzzy IF-THEN

obser-vations of n-input-single-output data pairs (X i, y i) (i=1, 2,

can be conveniently expressed using the following genericform

AND x 2 is A l

j2( )

AND,

…, x n is A l

jn( )

DESIGN OF ANFIS NETWORKS USING HYBRID GENETIC AND SVD METHODS FOR MODELING 169

way, the domains are appropriately selected so that all the

there exists A ( j) in equation (5) such that the degree of

Gaussian membership functions in the form

(6)

antecedents, respectively It is evident that that the number

of such parameters involved in the antecedents of ANFIS

dimension of input vector and r is the number of fuzzy sets

in each antecedent The fuzzy rule expressed in equation

It is evident that the input vector

impli-cation, the degree of such local fuzzy IF-THEN rule can be

evaluated in the form

(7)where

and

product inference engine and, finally, aggregating the

individual contributions of rules leads to the fuzzy system

in the form

equation (4) is available Equation (9) can be alternatively

represented in the following linear regression form

where D is the difference between f(X) and corresponding

actual output, y, and

(11)

It is therefore evident that equation (10) can be readily

pairs (X i, y i) (i=1, 2, …, m) in the form

(12)

rule such that the firing strength matrix P is obtained wheninput spaces are partitioned into a certain number of fuzzysets It is evident that the number of available training datapairs is usually larger than all the coefficients in theconclusion part of all TSK rules when the number of such

turns the equation (12) into a least squares estimation

can be expressed in the form

(13)Such modification of coefficients in the conclusion part ofTSK rules leads to better approximation of the data pairsgiven, in terms of minimisation of the difference vector D.However, such direct solution of normal equations is sus-ceptible to round-off error and, more importantly, to thesingularity of these equations

Therefore, in this paper, singular value decomposition isused as a powerful numerical technique to optimally deter-mine the linear coefficients embodied in the conclusionpart of the ANFIS model and deal with probable singulari-ties in equation (12) However, in this work, a hybridisation

of genetic algorithms and SVD is proposed to model therubber engine mount stiffness for the optimal design ofANFIS Such combination of genetic algorithms and SVDwere described in previous sections

3 APPLICATION OF GENETIC ALGORITHM (GA) TO ANFIS DESIGN

The incorporation of a genetic algorithm into the design ofsuch ANFIS models starts by representing the N(n+1) real-value parameters of {c j, σ j} as a string of concentrated sub-strings of binary digits Thus, each such sub-string repre-sents the fuzzy partitioning of antecedents of fuzzy rulesembodied in such ANFIS models in a binary coded form.The fitness (Φ) of each string of binary digits that repre-sents an ANFIS system, which models the engine mountstiffness, is readily evaluated in the form of where E is the

m S ≥ W= w [ 1 , w 2 , ,w s ] T

W= P ( T P ) 1 P T Y

170 J MARZBANRAD and A JAMALI

(14)objective function given by equation (3) and is minimised

through an evolutionary process by maximisation of

generating an initial population of binary strings, each

representing a candidate solution of the fuzzy

parti-tioning of the premise part of the rules Goldberg (1989)

stated that using the standard genetic operations of roulette

wheel selection, crossover, and mutation causes entire

populations of binary string to gradually improve

Simultane-ously, linear coefficients of the conclusion parts of TSK

rules corresponding to each chromosome representing the

fuzzy partitioning of the premise parts are optimally

deter-mined using SVD Therefore, ANFIS models of engine

mount stiffness that have progressively increasing fitness

are simultaneously determined by genetic algorithms and

SVD, respectively In other words, each chromosome that

represents the fuzzy partitioning of antecedents is related to

the corresponding linear coefficients of consequents

obtain-ed by the SVD method Figure 1 shows a schematic code

for this design process and a general ANFIS model

algo-rithm The following section provides a detailed summary

of the SVD application used to optimally determine the

linear coefficients in the linear equations

4 APPLICATION OF SINGULAR VALUE

DECOMPOSITION TO ANFIS DESIGN

In addition to the genetic information gained from the

antecedents of fuzzy sets involved in ANFIS networks,

singular value decomposition is also deployed for the optimal

design of consequents of such fuzzy systems Singular

value decomposition is the most common method used for

solving linear least squares problems, although some

singu-larities may exist in the normal equations The SVD of a

product of three matrices, a column-orthogonal matrix

elements (singular values), and an orthogonal matrix

such that

(15)Golub and Reinsch (1970) originally proposed the mostpopular technique for computing the SVD The optimalselection problem of W in equation (12) is first reduced to

which the reciprocals of zero or near zero singulars ing to a threshold) are set to zero Then, such optimal W areobtained using the following relation

For convenience, local coordinate axis shown in Figure

2 will be used throughout this paper There are sixgeometric parameters used to define the shape of the bushtype engine mounts, as shown in Figure 2 However, r i, r o

certain engine mount Therefore, three parameters, t r, t s and

t Z, are considered to be design variables

The stiffness values, derived from the system vibrationanalysis, are known as “dynamic” stiffness However, the

“static” stiffness can be simply computed by

(17)

stiffness, respectively, and η is a correction factor as stated

by Kim et al. (1992) The correction factor is generally inthe range of 1.2 to 1.6 In this paper, the static stiffness ismodelled, and, consequently, the dynamic stiffness can becomputed by a proper value of correction factor that can beselected for a particular rubber material

The static stiffness value is obtained from the quasistatic

DESIGN OF ANFIS NETWORKS USING HYBRID GENETIC AND SVD METHODS FOR MODELING 171

nonlinear finite element analysis by appropriate boundary

conditions included in the model Since the strain is relatively

small in this analysis, the classical Mooney-Rivlin (1992)

form of the strain energy is sufficient to describe the fully

incompressible hyperelastic material behaviour The

(18)where I 1 and I 2 are the first and second strain invariants,

respectively The coefficients C 10 and C 01 are determined

from the results of the unaxial tension test For the special

the FEA to obtain the stiffness in different directions The

ranges of design variables that were used in this study are

The finite element model (FEM) of an engine mount

with different dimensions, in ranges of design variables, is

built using ANSYS software A typical engine mount FEM

is shown in Figure 3 The structure is represented

quanti-tatively as finite collections of elements whose deformations

and stiffness can then be computed using linear algebraic

equations In these analyses, the fixed geometry values are

coefficients expected from stress-strain relationship have

been given as C 10=0.03622 and C 01=−0.00335

6 GENETIC/SVD BASED ANFIS MODELLING

AND PREDICTION OF RUBBER ENGINE

MOUNT STIFFNESS

In this paper, 36 different engine mount geometries were

considered in the FEA in order to obtain their stiffness in

3-orthogonal directions The stiffness values, together with

their corresponding geometry dimensions, were given by

In order to model a 3-input-single-output dataset, an

ANFIS with 2 linguistic terms in each antecedent,

equiva-lent to 2 Gaussian membership functions for each input

variable, was considered; that is, n=3 and r=2 It should be

noted that the number of parameters in each coefficient

vector in the conclusion part of each TSK-type fuzzy rule is

4, according to the assumed linear relationship of input

fuzzy rules were identified using the ANFIS given inMATLAB fuzzy-logic toolbox The corresponding fitness

is calculated as 1.022 and 1.125 for k xand k y, respectively

In order to demonstrate the effectiveness of the hybriddesign method, the genetic algorithm and singular valuedecomposition that was developed in this work was appliedfor the modelling of 3-input-single-output set of data ofengine mount for k x and k y The number of Gaussian member-ship functions for each input variable in the premise part ofthe rules was considered to be 2 During the evolutionaryprocess, the population size, mutation probability, crossoverprobability, and generation number were selected to be: 30,0.07, 0.7, and 150, respectively It should be noted that 4bits were chosen to be the binary representation of eachvariable, making the length of a chromosome 48 bits withrespect to 3×2×2=12 parameters Figures 4 and 5 show theperformance of different genetically obtained hybrid ANFISnetworks, each having a different number of rules in thecross-validation process for k x and k y, respectively It isevident from these figures that an ANFIS with only 7 rules(considering the exact value of fitness) is sufficient to

U=C 10 ( I 1 – 3 )+C 01 ( I 2 – 3 )

10 t ≤ ≤ r 15 25 t ≤ ≤ s 40 25 t ≤ ≤ z 40

Figure 3 Finite element model of an engine mount

Figure 4 Training errors of GA/SVD-designed ANFISwith different number of rules (kx modelling)

Figure 5 Training errors of GA/SVD-designed ANFISwith different number of rules (ky modelling)