International Journal of Automotive Technology, Vol 11, No 5, pp 611−616 (2010) DOI 10.1007/s12239−010−0073−6 Copyright © 2010 KSAE 1229−9138/2010/054−01 ENHANCEMENT OF NOx-PM TRADE-OFF IN A DIESEL ENGINE ADOPTING BIO-ETHANOL AND EGR S JUNG , M ISHIDA , S YAMAMOTO , H UEKI and D SAKAGUCHI 1)* 2) 2) 3) 2) Research and Development Department, Daihatsu Diesel Mfg Co Ltd., Shiga 524-0044, Japan Graduate School of Science and Technology, Nagasaki University, Nagasaki 852-8521, Japan Faculty of Engineering, Nagasaki University, Nagasaki 852-8521, Japan 1) 2) 3) (Received 12 December 2009; Revised 19 February 2010) ABSTRACT−For realizing a premixed charge compression ignition (PCCI) engine, the effects of bio-ethanol blend oil and exhaust gas recirculation (EGR) on PM-NOx trade-off have been investigated in a single cylinder direct injection diesel engine with the compression ratio of 17.8 In the present experiment, the ethanol blend ratio and the EGR ratio were varied focusing on ignition delay, premixed combustion, diffusive combustion, smoke, NOx and the thermal efficiency Very low levels of 1.5 [g/kWh] NOx and 0.02 [g/kWh] PM, which is close to the 2009 emission standards imposed on heavy duty diesel engines in Japan, were achieved without deterioration of the thermal efficiency in the PCCI engine operated with the 50% ethanol blend fuel and the EGR ratio of 0.2 It is found that this improvement can be achieved by formation of the premixed charge condition resulting from a longer ignition delay A marked increase in ignition delay is due to blending ethanol with low cetane number and large latent heat, and due to lowering in-cylinder gas temperature on compression stroke based on the EGR It is noticed that smoke can be reduced even by increasing the EGR ratio under a highly premixed condition KEY WORDS : PCCI engine, Bio-ethanol, EGR, PM-NOx trade-off INTRODUCTION of fuel injection to the onset of ignition on the basis of combustion tests with large quantities of cooled EGR and low cetane number fuels On the other hand, bio-ethanol as a carbon neutral fuel is one of the alternative fuels, and it is effective to reduce carbon dioxide emissions Kamio et al (2007) investigated the effect of ethanol fuels on HCCI-SI hybrid combustion using dual fuel injection Ishida et al (2004) showed the effectiveness of gasoline blended with gas oil in a diesel engine experimentally In this case, smoke was reduced markedly by blending gasoline having low cetane number and low evaporation temperature, resulting from a longer ignition delay The objective of the present study is to show one of the approaches for achieving low NOx and low PM combustion in a diesel engine by utilizing bio-ethanol The gas oil was blended with bio-ethanol for realizing a PCCI engine in the present experiment The effects of ethanol blend ratio and EGR ratio on ignition delay, premixed combustion, diffusive combustion, smoke density, concentrations of NOx, unburned hydrocarbon and carbon monoxide, and the thermal efficiency were investigated in detail Recent researches aiming at almost zero emissions on NOx and PM in diesel engines have shown that a homogeneous charge compression ignition (HCCI) engine or PCCI engine will be a promising way to accomplish the target In order to achieve the severe emission standards imposed on diesel engines such as the post new-long term emission standards for 2009 in Japan, the EURO VI standards for 2012 and the US standards for 2010, it is necessary not only to depend upon the aftertreatment system with NOx catalyst and PM catalyst but also to suppress NOx and PM formations in spray combustion process by forming the PCCI or HCCI condition The HCCI engine showed some problems on the limited operation range due to misfire and knock (Jung et al., 2005; Ishida et al., 2006, 2007, 2008; Jung et al., 2007), then, the PCCI engine is thought to be better than the HCCI engine in order to expand the engine load range In the recent studies on a PCCI engine operated with a high EGR and low cetane number fuel (Ogawa et al., 2007), and Li et al (2007, 2008) showed that smoke emission decreased with a longer ignition delay due to the low temperature combustion under the ultra-high EGR condition, and smoke was dependent strongly upon the premixing time from the end EXPERIMENTAL APPARATUS The test engine is a single cylinder high-speed naturally aspirated direct injection diesel engine, the type NFD 170(E) manufactured by YANMAR Co., Ltd The bore is 102 *Corresponding author e-mail: sukho1001@hotmail.com 611 612 Table Properties of test fuels Gas oil Gas oil (vol%) 100 Ethanol (vol%) Octanol (vol%) Lower heating value (MJ/kg) 42.9 Latent heat (MJ/kg) 0.25 Boiling point (ºC) 190-350 Oxygen (wt%) Cetane number 55 S JUNG et al EtOH30 EtOH50 68 48 29 48 38.3 35.2 0.43 0.55 − 10.1 27 − 16.8 18 mm, the stroke is 105 mm, and the compression ratio is ε=17.8 Test fuels are gas oil having cetane number of 55 and ethanol with that of about Table shows properties of tested fuels The blend fuel named EtOH30 consists of 68 vol% gas oil, 29 vol% ethanol with vol% octanol as a surface-active agent, and EtOH50 consists of 48% gas oil, 48% ethanol and 4% octanol The lower heating values of gas oil and ethanol are 42.9 and 26.8 MJ/kg respectively Ethanol shows a lower cetane number, lower evaporation temperature and larger latent heat compared with gas oil Figure shows the experimental test system The combustion tests were carried out under the conditions of a constant engine speed of 1,200±5 rpm, a constant suction air pressure of 0.1013 MPa at the intake manifold and a constant intake temperature of TIN =40oC even in the case with EGR, in other words, the cooled EGR EGR gas was charged into the mixing chamber located at 1,800 mm upstream of the intake manifold, and the fuel was injected into the combustion chamber directly at a constant injec- Figure Effects of ethanol and EGR on combustion timehistory (Pme=0.51 MPa) tion timing of 5o CA BTDC CO2 concentration% was measured at both intake and exhaust manifolds respectively to calculate the EGR ratio XEGR In the exhaust gas analysis, exhaust gas temperature Te o C, concentrations of carbon monoxide CO ppm, total unburned hydrocarbon THC ppm and nitrogen oxides NOx ppm, and smoke density were measured respectively as shown in Figure later The time-history of in-cylinder pressure was measured using the piezo type sensor and this output was sampled every one-fourth degree in crank angle by means of the channel combustion analyzer CB-467 manufactured by Ono Sokki Co Ltd The time-history of combustion pressure was the ensemble average sampled over continuous 350 engine cycles The data were transmitted to the personal computer and recorded on hard disks RESULT AND DISCUSSION Figure Combustion test system 3.1 Effects of Ethanol Blend Ratio and EGR Ratio on Combustion Time-history and Engine Performance Figure shows a change in time-history of combustion due to ethanol blend and EGR as well under the high engine load of Pme=0.51 MPa CA of the abscissa denotes the crank angle degree, and P, dQ/dθ and Lift in the ordinate denote a measured in-cylinder pressure, apparent heat release rate and needle valve lift respectively In the experiment, the EGR ratio was increased while the intake temperature was kept constant As the ethanol blend ratio increases and also the EGR ratio increases, ignition timing is retarded markedly, then, the premixed combustion becomes larger and the diffusive combustion decreases remarkably Figure shows changes in exhaust emissions and fuel ENHANCEMENT OF NOx-PM TRADE-OFF IN A DIESEL ENGINE ADOPTING BIO-ETHANOL AND EGR 613 Figure Definitions of ignition timing, ignition delay and diffusive combustion Figure Effects of ethanol and EGR on exhaust emissions and engine performance smoke without EGR (Smoke0) to smoke with EGR in each condition It should be noticed that, only in the case of EtOH50, smoke was reduced by increasing the EGR ratio The reason why smoke is reduced by EGR will be clearly shown in the latter section of this paper consumption due to engine load, where the parameters are fuels of gas oil and 50% ethanol blend fuel (EtOH50) and the EGR ratio The brake specific fuel consumptions reduced by lower heating value of each fuel are almost the same between these three cases The one of EtOH50 is a little lower than that of gas oil alone in the high load range NOx increases by means of ethanol blending because the premixed combustion increases due to longer ignition delay and due to lower evaporation temperature of ethanol, on the other hand, it is reduced markedly by EGR as shown in Figure Smoke is reduced remarkably in the case of 50% ethanol blend fuel, in addition, it is decreased by EGR further Figure shows smoke change rate due to EGR ratio Smoke/Smoke0 in the ordinate denotes a proportion of 3.2 Ignition Delay Due to Ethanol and EGR Figure shows the definitions of a ignition timing, ignition delay and approximated diffusive combustion curve of the heat release rate The ignition timing was defined as a zerocross point of the dotted line tangential to the heat release rate curve in a initial premixed combustion stage The ignition delay was defined as the crank angles between the start of injection and the ignition point The heat release rate curve during the diffusive combustion period was approximated by the Wiebe’s function (Miyamoto , 1985) as shown in Figure 5, and the cumulative heat release of diffusive combustion Qd was calculated by integrating the Wiebe’s function Figure shows change in the ignition delay due to the EGR ratio; where parameters are fuel and engine load The ignition delay is dependent largely upon the fuel cetane number; cetane numbers of the fuels vary from 55 of gas oil to 27 of the fuel EtOH30 and 18 of the fuel EtOH50 In Figure Change in smoke due to EGR Figure Effects of ethanol and EGR on ignition delay et al 614 S JUNG et al Figure Change in cumulative heat release during diffusive combustion Qd due to ignition delay the ethanol blend fuels, ignition delay is increased mainly by the lower cetane number and secondarily by the larger latent heat The smaller the cetane number is, the longer the ignition delay is The ignition delay increases markedly with increase in the ethanol blend ratio, and also increases further by increasing the EGR ratio However, variation of ignition delay due to engine load is relatively small except for the cases with the high EGR ratio Increase in ignition delay due to EGR is based on lowering the in-cylinder gas temperature on compression stroke The longer ignition delay promotes premixing between the fuel and intake charge Figure shows change in the cumulative heat release during diffusive combustion (Qd) due to ignition delay, where the parameter is the engine load The data in the figure include the cases with different ethanol blend ratio and the cases with different EGR ratio Qd, which seems to be a main factor of smoke emission, decreases almost linearly with increase in ignition delay at any engine load although Qd is larger at higher load because of larger fuel injection quantity 3.3 Relationship between PM and Diffusive Combustion The mass rate of diffusive combustion md was calculated by the following equation; (1) md = Qd /Hu where Hu is the lower heating value of the fuel Assuming that higher evaporation temperature components of gas oil burns in the diffusive combustion process, the lower heating value of gas oil is applied for Hu because ethanol has fairly lower evaporation temperature compared with gas oil, and it burns in the premixed combustion stage Figure shows a correlation between the mass rate of diffusive combustion md and the injection quantity of gas oil All Qd data shown in Figure are plotted again in Figure It is clear that the mass rate md correlates well with the injection quantity of gas oil, in other words, md is strongly dependent upon injection quantity of gas oil and Figure Corelation between mass rate of diffusive combustion md and injection quantity of gas oil Figure Relationship between PM and md/(1−XEGR) its amount is about 70% of the injected gas oil Figure shows the relationship between the particulates matter in the exhaust gas and the parameter md /(1−XEGR) for accessing the effect of md and EGR on reduction in PM; where 1−XEGR of the denominator is adopted instead of XEGR because the denominator is at XEGR =0 It is clear from Figure that the PM is strongly dominated by md /(1− XEGR) If the EGR ratio is constant, the PM increases as the mass rate of diffusive combustion md increases, and also the PM increases as the EGR ratio increases if md is constant It cannot be simply determined whether the PM decreases or increases because variation of md is dependent strongly upon ignition delay In order to reduce the PM by increasing the EGR ratio, the value of md /(1−XEGR) should be decreased as shown in Figure by two solid circles on the correlation line This condition is written by the following equation; md/( – XEGR ) D × 1.5 If the above condition is true, we can assume the image has an abnormal intensity variation caused by a shadow cast over the plate In this case, Niblack’s method is used for binarization Otherwise, a normal lighting condition is assumed, and Otsu’s method is applied Figure 5(a) shows two sample images with a shadowed region, and (b) and (c) ver hor ver hor hor ver LICENSE PLATE LOCATION METHOD UNAFFECTED BY VARIATION IN SIZE AND ASPECT RATIO 755 Figure Result of a license plate location Figure Two sample images with a shadowed region: (a) extracted LP candidate images; (b) results of Otsu’s method; (c) results of Niblack’s method; (d) and (e) similar-sized components extracted from the binary images of (b) and (c), respectively show the results of binarization by Otsu’s method and Niblack’s method, respectively If the intensity value of a pixel is greater than the threshold acquired from Otsu’s or Niblack’s method, then the pixel is converted to a white pixel Otherwise, the pixel is converted to a black pixel Thus, the white characters on a green plate are set to white pixels and the black characters on a white plate are set to black pixels Because a unified representation is convenient for the subsequent process, it is necessary to estimate the background color of a license plate We decide the background color by using the ratio of white pixels to the black pixels Figure 5(d) and (e) show the unified binarization results and the connected components of characters in images (b) and (c), respectively Along the five rows traversing the central part of the binarized region, we found zero crossings Zero crossing means that there is a fluctuation between the white pixels and the black pixels on the horizontal line If the number of zero crossings at the central line and the lower two lines is not sufficiently large, the upper two lines are further inspected The connected components are extracted along the horizontal line where a maximum zero crossing occurs In general, a plate has at least four similar-sized components, and the components are located horizontally This arrangement information is enough to decide whether an LP candidate contains a license plate or not The arrangement pattern is analyzed by the size of connected components and by the vertical overlap or the horizontal distance between two adjacent components A poor binarization result often causes failure in verifying the LP candidates Although the binarization results shown in the first column of Figure 5(b) are not good, four similar-sized components were successfully extracted In the case of the second column in Figure 5(b), we could not extract similar-sized components Indeed, as shown in Figure 5(e), the components were successfully extracted in the two binary images acquired using Niblack’s method Sometimes similar-sized components are not detected sufficiently due to broken strokes or unwanted connections with other components In this case, a missed or merged component is assumed, and its position is estimated using the location and the aspect ratio of the adjacent components If more than two similar-sized components are detected or assumed, the LP candidate having the most probable component pattern is selected as the region of a license plate Otherwise, it is assumed that none of the LP candidates contain a license plate The LP candidates in Figure appear as five binarized regions in Figure The topmost region has two similar-sized components, and the bottommost region has four similar-sized components The bottommost region is identified as a license plate region, which is displayed as a bold-line rectangle in Figure EXPERIMENTAL RESULTS We implemented the proposed method using C++ on a Pentium PC (2.2 GHz) to measure its performance Most of the previous studies have experimented with their own vehicle image database, so it is not reasonable to estimate the performance of each method by comparing the results of their experiments Thus, we have used not only our data set but also a public database introduced in Anagnostopoulos (2008) This is, as far as we know, the first benchmark study that experimented with a public license plate database Three sets of data were used in this experiment Set consisted of 319 color images captured on our campus We took photographs of parked cars at random, so the percentage (20.7%) of white plates may be considered to be the amount of newly registered cars since 2006 In Set 1, the resolution of 200 images was 1024×768 and that of the rest was 640×480 Sets and were acquired from the public database The database had 12 groups of vehicle images The ‘day_color(small_sample)’ group was selected as Set and the ‘day_color(large_ sample)’ group as Set Set had 67 parked vehicle images and its resolution was 800×600 Set consisted of 135 images with a mix of et al 756 M.-K KIM 640×480, 800×600, and 1792×1312 resolutions To compare the performance of the proposed method, we implemented two related methods (Bai et al., 2003; Wu et al., 2006) The experimental conditions used in this study differed from those of the two related studies, and all the parameter values used in the two related studies were not exactly known Thus, we used the same conditions to construct LP candidates The conditions used were as follows: Condition 1: WImage /10 < WMBR < WImage /2 Condition 2: HImage /30 < HMBR Condition 3: 1.5 < AspectRatio < 10.0 Condition 4: 402 < Area Condition 5: 0.4 < Density WImage and HImage are the width and height of an image To reduce processing time, images having a horizontal resolution above 1,000 were first reduced in size by half, so WImage and HImage were halved in this case WMBR and HMBR are the width and height of the MBR of an LP candidate, and AspectRatio is the ratio of WMBR to HMBR Density is Area/ (WMBR×HMBR), where Area is the number of foreground pixels in the MBR Benchmark results are shown in Table The first column in each set shows the average number of LP candidates detected, and the second column represents the rate of correct location We used two criteria in calculating the rate of correct location If we used the criterion raten, a correct location means that a license plate existed in one of the LP candidates In the case of rate1, it means that a license plate was correctly located The reason we used two criteria is that the two related papers did not provide sufficient information to verify the LP candidates Thus, the rate1 was computed in the proposed method only The proposed method outperformed the other two methods not only in locating Korean license plates in Set but also in locating Table Benchmark results of the proposed and comparison methods Set Set Set raten No raten No raten No (rate (rate1) (rate1) 1) Bai’s method 2.4 75.5% 2.2 70.1% 1.9 93.4% Wu’s method 1.4 82.8% 1.2 83.6% 1.4 86.1% 96.9% 2.3 94.0% 2.2 99.3% Proposed 3.2 (95.0%) (94.9%) (88.1%) method Table Benchmark results summarized as the license plate types Short Type Long Type All rate (rate1) rate (rate1) rate (rate1) Bai’s method 71.3% 87.1% 79.5% Wu’s method 86.5% 81.3% 83.7% 97.2% 97.1% 97.1% Proposed method (94.8%) (93.4%) (94.1%) n n n Figure Two sample vehicle images: (a) input images; (b)-(d) edge regions constructed by Bai’s, Wu’s, and the proposed methods, respectively; (e) license plate regions located by Wu’s method and Bai’s method; (f) Each result shows one license plate region located by the proposed method LICENSE PLATE LOCATION METHOD UNAFFECTED BY VARIATION IN SIZE AND ASPECT RATIO Table Comparison of computational times (unit: ms) 1024 640 800 1792 Average ×768 ×480 ×600 ×1312 of all Bai’s method 185 287 439 540 363 Wu’s method 107 164 247 301 205 Proposed method 197 305 465 564 383 Greek license plates in Set and Set If we classify the license plates according to the number of characters N, the license plates can be divided into short type (N < 7) and long type (N ≥ 7) Table shows the results summarized with the view of the license plate types As might be expected, the results showed that the proposed method was superior to the two comparison methods regardless of the license plate types The two methods (Bai et al., 2003 Wu et al., 2006) failed in the case where vertical edges in a plate region were merged with relatively long vertical edges outside of the plate region as shown in Figure 7(b1) The two methods were also weak because all of the edge regions located in a plate region were not merged into a single region In contrast, the failures of our method were caused by an improper binarization that made it difficult to extract the components correctly A high skew of license plates was a common cause of failure of all three methods Figure shows edge regions, LP candidates, and detected regions of license plates Figure 7(b)~(d) shows the edge regions constructed by Bai’s method, Wu’s method, and the proposed method In Figure 7-(b1) and (c2), the edge regions are touched by long vertical edges and neighboring edges outside of plate regions The touching did not allow us to extract a license plate Wu’s method succeeded only in the case of input image (a1) and Bai’s method succeeded only in the case of input image (a2) The extracted LP candidates are shown in Figure 7(e1) and (e2) One of the two MBRs contains a license plate, and these extraction results are correct if the criterion raten is applied Only the proposed method succeeded in both cases The results are shown at Figure 7(f1) and (f2), in which LP candidates and the regions of license plates are represented by rectangles Computation time is also an important factor to be considered for developing a practical system The average computational times of the three methods are shown in Table Because the images of 1024×768 and 1792×1312 resolution were down sampled, the column sequence is the order of image size Wu’s method was the fastest of the three methods The proposed method was slower than either of the other two methods, but it was fast enough for applications such as watching illegally parked vehicles and finding missing vehicles, etc CONCLUSIONS This study proposed a new license plate location method 757 that is unaffected by the variations in size and aspect ratio The method solved some of the problems of the previous methods First, most of the unwanted merging problems caused by long vertical edges were solved by removing the long vertical edges beforehand Second, license plates that have multiple colors and aspect ratios could be effectively extracted by using an edge region Finally, the performance of verifying LP candidates improved due to the arrangement pattern of the connected components The experimental results showed that the proposed method was effective regardless of the license plate types Nonetheless, some faults occurred when components were touched by other components and a license plate was skewed The main cause of the touching problem was an improper binarization The selective binarization method was more effective than Otsu’s and Niblack’s methods, but it still can be improved Our future work will be dedicated to solving the problems caused by skew and shadows cast on the plate ACKNOWLEDGEMENT−We would like to thank E Kayafas and his members at the multimedia technology laboratory in National Technical University of Athens for providing their public license plate image database REFERENCES Anagnostopolus, E C., Anagnostopolus, E I., Psoroulas, D I., Loumos, V and Kayafas, E (2008) License plate recognition from still images and video sequences: A survey IEEE Trans Intelligent Transportation Systems 9, 3, 377−391 Bai, H., Zhu, J and Liu, C (2003) A fast license plate extraction method on complex background Proc Intelligent Transportation Systems, 2, 985−987 Chang, S.-L., Chen, L.-S., Chung, Y.-C and Chen, S.-W (2004) Automatic license plate recognition IEEE Trans Intelligent Transportation System 5, 1, 42−53 Gao, D.-S and Zhou, J (2000) Car license plates detection from complex scene Proc Signal Processing, 2, 1409− 1414 Guo, J.-M and Liu, Y.-F (2008) License plate localization and character segmentation with feedback self-learning and hybrid binarization techniques IEEE Trans Vehicular Technology 57, 3, 1417−1424 Jia, W., Zhang, H and He, X (2007) Region-based license plate detection J Network and Computer Applications, 30, 1324−1333 Kim, K I., Jung, K and Kim, J H (2002) Color texturebased object detection: An application to license plate localization LNCS 2388, 293−309 Kong, J., Liu, X., Lu, Y and Zhou, X (2005) A novel license plate location method based on textural feature analysis Proc Int Symp Signal Processing and Information Technology, 275−279 Li, G., Zeng, R and Lin, L (2006) Research on vehicle 758 M.-K KIM license plate location based on neural networks Proc based on histogramming and mathematical morphology Mahini, H., Kasaei, S., Dorri, F and Dorri, F (2006) An efficient features-based license plate localization method Proc ICPR'06, 2, 841−844 Niblack, W (1986) An Introduction to Digital Image Processing Prentice Hall Englewood Cliffs/N.J Otsu, N (1979) A threshold selection method from graylevel histogram IEEE Trans Systems, Man, and Cybernetics 9, 1, 62−66 Shapiro, V., Gluhchev, G and Dimov, D (2006) Towards a multinational car license plate recognition system Machine Vision and Applications, 17, 173−183 Wu, H P., Chen, H.-H., Wu, R.-J and Shen, D.-F (2006) License plate extraction in low resolution video Proc ICPR'06, 1, 824−827 Yang, F and Ma, Z (2005) Vehicle license plate location Yang, Y.-Q., Bai, J., Tian, R.-L and Liu, N (2005) A vehicle license plate recognition system based on fixed color collocation Proc 4th Int Conf Machine Learning and Cybernetics, 9, 5394−5397 Youssef, M S and Abdelrahman, B S (2008) A smart access control using an efficient license plate location and recognition approach Expert System with Application, 34, 256−265 Zhang, H., Jia, W., He, X and Wu, Q (2006) Learningbased license plate detection using global and local features Proc 18th ICPR'06, 2, 1102−1105 Zheng, D., Zhao, Y and Wang, J (2005) An efficient method of license plate location Pattern Recognition Letters, 26, 2431−2438 ICICIC’06, 174−177 Proc 4th IEEE Workshop on Automatic Identification Advanced Technologies, 89−94 Copyright © 2010 KSAE 1229−9138/2010/054−18 International Journal of Automotive Technology, Vol 11, No 5, pp 759−765 (2010) DOI 10.1007/s12239−010−0090−5 VIBRATION TRANSMISSION REDUCTION FROM A CENTRIFUGAL TURBO BLOWER IN A FUEL CELL ELECTRIC VEHICLE * Y S KIM, E Y KIM, Y W SHIN and S K LEE Department of Mechanical Engineering, Inha University, Incheon 402-751, Korea (Received December 2009; Revised 22 February 2010) ABSTRACT−This paper presents a multi-body flexible dynamic analysis of a centrifugal turbo blower for a fuel cell electric vehicle (FCEV) based on the application of computer-aided engineering (CAE) to predict the acceleration at the mount position of the blower This predicted acceleration is validated by using the measured acceleration data The numerical simulation for the multi-body flexible dynamics of the blower is used not only to identify the most effective mount among four mounts in an FCEV by controlling the complex stiffness of the isolator, but also to suggest the range of complex stiffness of the isolator at the most effective mount This numerical simulation technology can be useful for the estimation of the variation of vibration transmission for the structural modification of the turbo blower KEY WORDS : FCEV, Turbo blower, CAE, Dynamic analysis, Vibration INTRODUCTION With the recent ballooning oil prices and consumers’ increasing concerns about the environment, an automobile is required that has high fuel efficiency and uses ecofriendly fuel for an internal combustion engine vehicle (ICEV) To meet these demands, automakers are developing automobiles using alternative energy Among these, a fuel cell electric vehicle (FCEV) is introduced as a practical alternative because it uses hydrogen as a fuel and releases only oxygen and water after a chemical reaction (Yang, 2000) An FCEV is a type of electric vehicle that is driven by an electric motor, as shown in Figure The stack in an FCEV is the most important part for generating electricity and it requires hydrogen and air with a certain level of humidity Many types of equipment are used for the continuous air supplement to the stack, but a centrifugal turbo blower has primarily been used (Larminie and Dicks, 2003) It is known that the blower is one of the dominant vibration and noise sources in an FCEV (Vielstich et al., 2003) because the centrifugal turbo blower operates at very high speed, above 30,000 rpm, in order to increase the pressure of the air, which should be supplied to the stack of an FCEV, using the rotation of its impeller blades Therefore, the noise vibration and harshness (NVH) in an FCEV are influenced by the noise and vibration originating from the blower The vibration of the blower is generated by an unbalance of mechanical components, rotating at high speed The vibration is transmitted to the chassis frame through vibration isolators (Goodwin, 1989; Genta, 1999; Figure Major components of a fuel cell electric vehicle Xu, and Marangoni, 1994; Harsha, 2006; Lee, 2000) In this paper, simulation technology is developed to predict the vibration generated from the blower based on CAE and the rotor dynamics This technology is also used to suggest the most effective among four mounts to reduce the vibration transmission of the blower and the optimal dynamic complex stiffness of the vibration isolator at that mount In order to develop this technology, finite element (FE) modeling and dynamic analysis of the blower are performed by using multi-body dynamic analysis software In this process, the measured and the simulated accelerations at either side of the vibration isolator on each of the four mounts are compared to validate the simulation results The simulated accelerations are used to evaluate the effectiveness of vibration isolators and to determine the optimal complex stiffness of the vibration isolator *Corresponding author e-mail: sangkwon@inha.ac.kr 759 760 Y S KIM, E Y KIM, Y W SHIN and S K LEE MODELING OF THE BLOWER FOR DYNAMIC ANALYSIS A simulation model for dynamic analysis of the blower is required to predict the acceleration at each mount of the blower Rigid body models and finite element models of the blower components are made by using CATIA (PLM Solution) geometry files of the blower, and these models are combined through ADAMS (MSC software Co.) 2.1 Geometric Modeling The blower is composed of many components, as shown in Figure The rotating shaft and the splitter impeller are connected to the end of the rotating shaft Bearings are also connected to each end of the rotating shaft A guide vane for increasing air compression efficiency is connected to the blower body through the motor housing, brackets, and so on CATIA geometry files of the blower components are converted to parasolid format by using MSC.Nastran, and these format files are then imported into MSC.ADAMS Each converted component is modeled as a rigid body that has mass, the center of mass and the moment of inertia Figure shows the important components and complete geometry of the blower 2.2 Finite Element Model of the Blower The mesh production is performed for finite element analysis by using the geometry model of the blower with MSC.Patran Rigid body element (RBE2) is used to connect each component with a bolt connection Through finite element analysis using MSC.Nastran, the modal parameters such as mode shape, natural frequency and modal vector are calculated These results are compared with those of an experimental modal analysis (EMA) to confirm the accuracy of the FE model A modal assurance criterion analysis (MAC) is generally used for this validation MAC analysis is a method of evaluating the orthogonal property for natural vectors of the vibration system Validated components are converted to modal neutral file Figure Geometry model of the blower Figure Finite element model of the blower (MNF) format, and then these models are imported to MSC.ADAMS for the dynamic analysis Figure shows a finite element model of the blower 2.3 Modeling of the Blower Component Rigid body models and finite element models are connected in MSC.ADAMS The motion of each component is appropriately constrained 2.3.1 Connections of the blower components Coupled components like the rotating shaft and the impeller are connected by using a fixed joint Components having only rotating motion are connected by using a revolute joint In order to rotate the shaft and the impeller, the torque measured in the laboratory is applied to the rotating shaft A bushing joint is used for the connection of the blower bracket with the chassis frame (Kim and Lee, 2008) Figure shows various joint configurations 2.3.2 Vibration isolator mounts and bearing connection The isolator rubber between the upper bracket and the Figure Joints of the blower components and the torque excitation method VIBRATION TRANSMISSION REDUCTION FROM A CENTRIFUGAL TURBO BLOWER 761 THEORY FOR THE SHAFT AND BEARING MOTION Figure Modeling of vibration isolator at mount points and bearing contact Blower vibration is generated by an unbalance of mechanical components, the rotation of bearings at high speed and the rotating asymmetry These vibrations are transmitted to the chassis frame through the vibration isolator at the mounts In this section, a theoretical analysis of the vibration of the rotating shaft in the blower is presented These results are used for the dynamic simulation for the blower vibration 3.1 Jeffcott Rotor Model The Jeffcott rotor model consists of a flexible, massless and uniform shaft with a flat disk illustrated as shown in Figure If the center of rotation is not coincident with the center of gravity by imbalance, the relations between the two centers are represented as follows (1) Figure Boundary conditions for the blower lower bracket at four mounts of the blower has nonlinear characteristics Therefore, the isolator rubbers are modeled by using nonlinear bushing joints in which the characteristics of the rubber are represented by the stiffness coefficient for deformation and the damping coefficient for the deformation velocity The bearings are modeled using Hertz’s contact theory (Johnson, 1987) Theoretical analysis of the bearing will be explained in the next section Figure shows the modeling of the vibration isolator at the mount points and the bearing modeled by contact theory Figure shows the boundary conditions of the blower model Figure shows a model of the assembled blower for dynamic analysis Figure Assembled model of the blower for dynamic analysis The equation of motion is represented in equation (2) by Newton’s law of motion (2) By substituting equation (1) into equation (2) (3) If the system is at steady state, equation (3) becomes the following: Figure Model and coordinate of a Jeffcott rotor 762 Y S KIM, E Y KIM, Y W SHIN and S K LEE (4) Using the complex notation, equation (4) is expressed as equation (5): (5) In equation (5), the eccentricity between the center of rotation and the center of gravity causes the unbalance It has an effect on the rotating orbit of the shaft, which rotates with the same speed as a shaft having an effective radius Therefore, this force is called the forward synchronous excitation The unbalance response of the system is given by and the critical speed with circular whirling motion is given by these expressions: (10) If the complex whirl speed (λ ) is the same as λ = −ω in ξη coordinates, the dimensionless whirling speed will be expressed as follows: ' ' (11) where, α* = η / ξ is the stiffness ratio If the stiffness ratio is 1, the secondary critical speed is given by: k k (6) (12) where ω* = ω /ω is a dimensionless rotating speed The unbalance response has the same speed and phase as the unbalance force Therefore, the response is called the forward synchronous circular whirling If the rotating speed is equal to the critical speed, the vibration approaches infinity The system with rotating asymmetry has not only whirling motion by the unbalance but also whirling motion with 2x rotating speed (Muszynka, 2005) cr 3.2 Rotating Asymmetry If the cross section of the shaft is not uniform or the shaft has defects, the rotating asymmetry will be caused by the non-isotropic stiffness of the shaft The reaction force of the system is expressed in equation (7), using a reference coordinate with rotation Figure shows the reference coordinates of the system: ' 3.3 Rolling Element Bearing A bearing with rolling elements causes nonlinear motion within the system by Hertzian contact forces, clearance and waviness of the bearing Figure 10 shows a bearing model with rolling elements Using Hertz’s theory, the elastic deformation, the contact force and the stiffness of bearing, respectively, are expressed as follows (Brändlein , 1999): et al (7) If damping is negligible, the equation of motion is given by the following: (8) (13) The system’s equation of motion of the bearing with rolling elements (n: number of rolling elements) is (Harsha, 2006): If the system is at steady-state, the unbalance response is written as follows: (9) Figure Reference frames with rotating asymmetry Figure 10 Model of a rolling element bearing VIBRATION TRANSMISSION REDUCTION FROM A CENTRIFUGAL TURBO BLOWER (14) Equation (14) shows the second order nonlinear differential equation of the bearing with rolling elements Deformation of the bearing inner race and the partial differential equation caused by the deformation is given by: 763 Table Parameters for the dynamic analysis of the simulated blower Dynamic parameter Specification Mass of the impeller (m) 0.73 kg Mass of the bearing rolling element (m ) 0.045 kg 0.078 kg Mass of the bearing inner race (m ) 0.078 kg Mass of the bearing outer race (m ) Diameter of the bearing rolling element (d ) mm Length over which the rollers are actually 7.344 mm in contact (l ) Internal radial clearance (δ ) 10 µm Mass eccentricity of the impeller imbal0.03 mm ance (ε) j in out j eff (15) This deformation affects the vibration of the blower 3.4 Simulated Acceleration The simulated accelerations at either bracket, as shown in Figure 6, are predicted by the dynamic analysis of the assembled model of the blower, as shown in Figure The point mass is attached to the rotating impeller of the blower in the simulation model to give the eccentricity between the center of mass and the center of gravity The rotating shaft is accelerated from to 30,000 rpm Table shows the parameters for the dynamic analysis of the blower The left Figure in Figure 11 shows the assembled model of the blower and the right Figure shows the time history of the simulated accelerations at either bracket of the blower The upper time history is related to the acceleration of the blower side bracket at mount number of the blower, and the lower time history is related to the acceleration of the frame side bracket Figure 12 shows the waterfall analysis for the time history of the simulated acceleration as shown in Figure 11 According to these waterfall analysis results, there are many harmonic components of the rotating shaft that comprise the unbalance response This is caused by the Figure 12 Waterfall analysis for the simulated accelerations unbalance force MODIFICATION OF ISOLATION STIFFNESS FOR VIBRATION REDUCTION Figure 11 Assembled simulation model of the blower and the simulated accelerations at the brackets on either side In order to validate the numerical simulation for the dynamic analysis based on the CAE technology, which uses the theoretical analysis discussed in section 3, the acceleration at each mount is measured and its results are compared with simulated accelerations The confirmed simulation is used to the modify the stiffness of the isolator for the vibration reduction of the blower 764 Y S KIM, E Y KIM, Y W SHIN and S K LEE Figure 13 Accelerometers attached to the mount of the blower 4.1 Baseline Test of the Blower Vibration For the vibration test, three directional accelerometers are attached to the upper and the lower brackets of the blower mount Figure 13 shows the accelerometers attached to either bracket of the blower There are four mounts on the centrifugal turbo blower The blower is attached to a heavy structural frame by these mounts in the laboratory During the rotation, the shaft of the blower is accelerated to 30,000 rpm and the three directional accelerations at these four mounts are measured Figure 14 shows the overall acceleration level measured at either bracket of the number mount According to these results, the acceleration level is attenuated by the vibration isolator of the number mount The attenuated level of the acceleration is approximately 20 dB These accelerations are used for the validation of the simulated acceleration based on the CAE technology discussed in section 4.2 Validation for the Dynamic Analysis of the Blower In order to validate the simulated acceleration obtained by using the simulation model based on CAE technology, the measured acceleration is compared with the simulated acceleration, as shown in Figure 15 According to these results, the simulated accelerations compare well with the measured accelerations Some differences are due to hydraulic pressure and modeling error Hydraulic pressure is the major source of error because the blower inhales air at very high speed and Figure 15 Comparison between the measured and simulated accelerations of the blower exhausts it at very high pressure Forces are decreased through connections between blower components The temperature of the fluid in the bearings and the friction force also have an effect on the error 4.3 Reduction of Vibration Transmission The function of the mount system is to support the blower weight and to isolate transmission of vibration from the blower to the frame The mount system consists of the bracket and isolator rubber The modification of the complex stiffness of the isolator rubber controls the rate of the vibration transmission In general, high frequency vibration can be reduced by adapting the low complex dynamic stiffness (Yu , 2001) However, excessively low stiffness causes substantial deformation, which becomes the cause of damage to the components of blower Therefore, it is necessary to simulate the vibration transmission, based on the simulation technology, in an advanced, low-cost experiment for the enhancement of the noise and vibration performance in an FCEV In this paper, the dynamic complex stiffness of the vibration isolator of the blower is modified using the simulation model to reduce the vibration transmission In order to this work, the root-mean-square (RMS) variation of the accelerations and the displacement at the frame side bracket, corresponding to the change of the rate of the stiffness, are predicted and their results are presented as et al Figure 14 Comparison between the blower side acceleration and the frame side acceleration at the number mount VIBRATION TRANSMISSION REDUCTION FROM A CENTRIFUGAL TURBO BLOWER 765 lation method was used to find the most effective mount to reduce the vibration transmission among four mounts from the blower to the frame It was found that mount number was the most effective mount such that a 30% reduction of the complex stiffness of the vibration isolator at mount number yielded a 22% decrement of vibration transmission at that mount A reduction of the stiffness over 30% does not have any effect on the reduction of vibration transmission ACKNOWLEGEMENT−This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MEST) (No 2010-0014260) REFERENCES Figure 16 RMS acceleration and displacement at the frame side bracket versus the change of the complex stiffness shown in Figure 16 Compressive and shear strains not exceed 20% and 30% of the length of the vibration isolator (Tohara, 1975) According to these results, the modification of the complex stiffness of the vibration isolator at the mount number is the most effective way to reduce vibration transmission The acceleration decreases until the stiffness is reduced by 30% The reduction of the stiffness over 30% does not have any effect on the reduction of vibration transmission The 30% reduction of the stiffness has the effect of a 22% reduction of the displacement CONCLUSION A centrifugal turbo blower is the primary vibration and noise source in a fuel cell electric vehicle, although it is also an important component for generating electric power for the FCEV This paper attempts to simulate the vibration generated from the blower based on the CAE technology and to apply the results to design the optimal stiffness of the vibration isolator at the mount of the blower To validate the simulated vibration, the vibrations on both sides of the vibration isolator bracket are measured in a laboratory experiment and the results are compared with the simulated vibration The measured vibrations corresponded very well with the simulated vibrations with only a small difference due to the modeling error This simu- Brändlein, J., Eschmann, P., Hasbargen, L and Weigand, K (1999) Ball and Roller Bearings Theory, Design and Application 3rd edn Wiley New York Genta, G (1999) Vibration of Structures and Machines Springer New York Goodwin, M J (1989) Dynamics of Rotor-Bearing Systems Unwin Hyman Boston Harsha, S P (2006) Nonlinear dynamic analysis of a highspeed rotor supported by rolling element bearing J Sound and Vibration , , 65−100 Johnson, K L (1987) Contact Mechanics Cambridge University Press Cambridge Kim, S J and Lee, S K (2008) Prediction of interior noise by excitation force of the powertrain based on hybrid transfer path analysis Int J Automotive Technology , , 577−583 Larminie, J and Dicks, A (2003) Fuel Cell Systems Explained 2nd edn Wiley New York Lee, S K (2000) Application of the vibrational power flow to a passenger car for reduction of interior noise Shock and Vibration , , 277−285 Muszy nka, H A (2005) Rotordynamics Taylor & Francis Boca Raton Tohara, H (1975) Rubber Vibration Isolator Japan Gendai Kogakusha Tokyo Vielstich, W., Lamm, A and Gasteiger, H A (2003) Hand290 1/2 ' book of Fuel Cells: Fundamentals, Technologies and Applications, 4, Wiley New York Xu, M and Marangoni, R D (1994) Vibration analysis of a motor-flexible coupling-rotor system subject to misalignment and unbalance: Part 1, 2, J Sound and Vibration , , 663−691 Yang, W.-C (2000) Fuel cell electric vehicles: Recent advances and challenges Int J Automotive Technology , , 9−16 Yu, Y., Naganathan, N G and Dukkipati, R V (2001) A literature review of automotive vehicle engine mounting system Mechanism and Machine Theory , , 123− 142 176 1 36