Trans. Japan Soc. Aero. Space Sci. Vol. 51, No. 173, pp. 146 150, 2008 Nacelle Chine Installation Based on Wind-Tunnel Test Using Efficient Global Optimization By Masahiro KANAZAKI, 1Þ Yuzuru YOKOKAWA, 1Þ Mitsuhiro MURAYAMA, 1Þ Takeshi ITO, 2Þ Shinkyu JEONG 3Þ and Kazuomi YAMAMOTO 1Þ 1Þ Aviation Program, Japan Aerospace Exploration Agency, Chofu, Japan 2Þ Institute of Aerospace Technology, Japan Aerospace Exploration Agency, Chofu, Japan 3Þ Institute of Fluid Science, Tohoku University, Sendai, Japan (Received July 2nd, 2007) Design exploration of a nacelle chine installation was carried out. The nacelle chine improves stall performance when deploying multi-element high-lift devices. This study proposes an efficient design process using a Kriging surrogate model to determine the nacelle chine installation point in wind-tunnel tests. The design exploration was conducted in a wind-tunnel using the JAXA high-lift aircraft model at the JAXA Large-scale Low-speed Wind Tunnel. The objective was to maximize the maximum lift. The chine installation points were designed on the engine nacelle in the axial and chord-wise direction, while the geometry of the chine was fixed. In the design process, efficient global optimization (EGO) which includes Kriging model and genetic algorithm (GA) was employed. This method makes it possible both to improve the accuracy of the response surface and to explore the global optimum efficiently. Detailed observations of flowfields using the Particle Image Velocimetry method confirmed the chine effect and design results. Key Words: High-lift System, Nacelle Chine, Kriging Model, Efficient Global Optimization (EGO) Nomenclature C L : lift coefficient C Lmax : maximum lift coefficient : ratio to nacelle length : angle to direction of surroundings around nacelle c nacelle : nacelle length X, Y, Z: scalar components of CAD data yðþ: function to be modeled x: design vector denoting position in design space E½IðÞŠ: expected improvement ^y: predicted value on model f max : maximum value among sample points s 2 : mean squared error of predictor : standard normal distribution : standard normal density 1. Introduction When an aircraft lands or takes off, its wing should have sufficient lift at low speeds and a high angle of attack. Under such conditions, a high-lift system that increases the lift coefficient at low speeds is required. Consequently, high-lift systems are key interest in aircraft design due to their impact on landing/take-off performance and payload capacity. 1) One design problem for high-lift systems is how to increase maximum lift at low speeds and high angle of attack. In aircraft with engine nacelles on the wing lower surface, the vortex from the nacelle pylon promotes flow Ó 2008 The Japan Society for Aeronautical and Space Sciences separation on the wing and reduces maximum lift. To avoid this separation, a nacelle chine, 2) which acts as a vortex generator as shown in Fig. 1, is used for transonic aircraft. The chine generates a counter-rotating vortex that counterbalances the vortex from the nacelle pylon. As a result, separation on the wing upper surface is suppressed and maximum lift is increased at high angle of attack (Fig. 2). Low-speed wind-tunnel tests for high-lift performance of civil aircraft were carried out at Japan Aerospace Exploration Agency (JAXA). These tests focused on the multielement high-lift wing using an experimental model called JAXA Standard Model (JSM) 3) consisting of fuselage, slat, main wing, flaps and engine nacelle (Fig. 3). Several tests were performed using force measurement, oil-flow observation, particle image velocimetry (PIV), and pressure sensitive paint (PSP). After these tests, nacelle chine installation was considered to establish a design method to improve high-lift performance. This research discusses in-board side nacelle chine installation using genetic algorithm (GA) on Kriging surrogate model. 4,5) The proposed method performs exploration using GA in a real-time wind-tunnel test. Many studies considering GA as a design optimizer 6 8) are based on computational evaluation, because GA has a large evaluation cost. To reduce this evaluation cost, this research introduces the Kriging surrogate model. In Kriging-based GA, expected improvement (EI) 4,5) values are used as criteria to select additional sample points. Here, this selection process is called Efficient Global Optimization (EGO). This method makes it possible to both improve the accuracy of the surrogate model ad to explore
Nov. 2008 M. KANAZAKI et al.: Nacelle Chine Installation Based on Wind-Tunnel Test Using Efficient Global Optimization 147 Fig. 1. Nacelle chine installed on in-board side of engine nacelle for landing configuration. Fig. 4. 6:5m 5:5mlow-speed wind-tunnel and JAXA standard model in the test section. Installation point Nacelle chine 30deg. 0deg. 90deg. 40%c nacelle 80%c nacelle 40% nacelle nacelle 90deg. c nacelle nacelle Fig. 2. Lift improvement when chine installed at correct position. Fig. 5. Design parameters. the global optimum efficiently. The nacelle chine installation design is performed by searching for optimum additional sample points based on EGO using a Distributed Genetic Algorithm (DGA). 10) After several sample installation points have been obtained, the optimum design can be determined. In this study, several points among final sample designs are selected and their flowfields are investigated in detail using PIV. 2. Evaluation The aerodynamic performance of sample installation points for the Kriging model was evaluated based on lift force measurements in low-speed wind tunnel tests. These tests were carried out in the 6:5m 5:5m low-speed wind tunnel (Fig. 4) at the Wind Tunnel Technology Center (WINTEC) of JAXA. The flow velocity was set at 60 m/s and measurement was performed while changing the angle of attack. 3. Formulation Fig. 3. JAXA standard model for high-lift system examination: top view, bottom view. 3.1. Design variables As shown in Fig. 5, the design parameters are and. Each is limited as follows:
148 Trans. Japan Soc. Aero. Space Sci. Vol. 51, No. 173 Table 1. Samples used for Kriging model. Outboard Inboard Inboard Design Range Fig. 6. Outboard Design Range Grid on nacelle of wind-tunnel test. Sample No. C Lmax 1 0.57 61.65 3.02 2 0.42 56.92 3.06 3 0.68 48.66 3.01 4 0.61 85.51 2.78 Initial 5 0.79 76.72 2.88 samples 6 0.56 71.72 2.93 7 0.66 36.95 3.07 8 0.74 80.67 2.86 9 0.52 31.19 3.05 10 0.46 47.68 3.05 ad1 0.78 30.00 3.01 ad2 0.57 43.00 3.06 Additional samples ad3 0.62 32.00 3.07 ad4 0.56 37.00 3.06 ad5 0.62 38.00 3.06 ad6 0.48 60.00 3.04 ad7 0.73 30.00 3.06 The design system procedure is shown in Fig. 7: 5) First, initial sample designs are evaluated and the Kriging surrogate model 1) is generated for the objective function, C Lmax. Next, the models are improved using additional sample points, because the model includes uncertainty. Additional samples are determined by maximizing the EI value. 4,5) The EI for the present maximization problem can be calculated from the Kriging model as follows: E½IðxÞŠ ¼ ð^y f max Þ ^y f max þ s ^y f max ð1þ s s Additional sample points are explored using DGA. 5) Robust exploration of the global optimum and improvement of the model can be achieved simultaneously, by selecting the maximum EI value as the additional sample point. 5. Results Fig. 7. Procedure of present design process. 0:4c nacelle 0:8c nacelle 30 ðdeg.þ 90 ðdeg.þ A grid is drawn on the nacelle as installation points as shown in Fig. 6, and the nacelle chine trailing edge is placed with reference to this grid. 3.2. Objective function The considered objective function is to maximize C Lmax. C L measured while changing the angle of attack from 0 deg. Then C Lmax is obtained from the C L - curve. 4. Overview of Present Design Method 5.1. Kriging model Ten sample designs were used to construct the Kriging model. To explore better solutions based on the EI maximization, the samples were added seven times (Table 1). Figure 8 shows C Lmax plots against -. The installation point where C Lmax increases can be selected by looking at this picture. Accordingly, if is set to around 60 deg., must be 0.4c nacelle 0:5c nacelle. In short, the chine should be installed near the leading edge of the nacelle. Conversely, if is set to around 30 deg. must be 0:6c nacelle 0:7c nacelle. This is a reasonable result, because the vortex from the nacelle chine will easily interfere with the vortex from the pylon when the chine is installed near the nacelle pylon. Moreover, is set to over 70 degree, C Lmax should be decreased for any value. 5.2. Projection of Kriging model to 3D CAD model For clearer understanding, the 2D C Lmax plot is projected onto a 3D CAD model of the nacelle as shown in Fig. 9. To
Nov. 2008 M. KANAZAKI et al.: Nacelle Chine Installation Based on Wind-Tunnel Test Using Efficient Global Optimization 149 Fig. 8. C Lmax plot predicted by Kriging model about. Fig. 10. Improvement of lift curve: nacelle chine installed at optimum point, comparison of lift curves. Fig. 9. C Lmax plot projected on CAD data of experimental model. generate this figure, the 3D vector (X, Y, Z) is translated into the 2D vector (, ). The Kriging model predicts C Lmax for this vector and the C Lmax contour is illustrated on the nacelle. The relationship between the nacelle chine and other components can be investigated by projecting the 2D plot on the 3D CAD data. 5.3. Improvement of lift curve Using the EGO process, sample No. 7 in Table 1 was selected as the optimum design (Fig. 10). Figure 10 compares the lift curve for sample No. 7 and without the chine configuration. From this figure, C Lmax is increased and the stall angle is delayed. Conversely, the presence of the chine has no impact on C L in the linear region. This suggests that the nacelle chine can increase C Lmax without negatively affecting landing performance. 5.4. Flowfield measurement Figure 11 show the velocity magnitude contour obtained by the PIV method with/without a chine at a 13 deg. angle of attack. Figure 11 shows the picture without a nacelle, and shows the picture with nacelle chine installed at the optimum point (sample No. 7). From Fig. 11, the vortex appears from the nacelle pylon, and interacts the flow on the upper wing surface. As a result, separation breaks out. The decrease in maximum lift is caused by this separation. On the other hand, according to Fig. 11, the vortex has weakened. Thus, the separation is not occurred by the vortex from the nacelle pylon and the maximum lift is increased. 6. Conclusions Efficient global optimization of the nacelle chine installation for landing was performed using the Kriging surrogate model. The objective function was the maximum lift coefficient for the landing configuration. The aerodynamic performance was evaluated based on wind-tunnel tests. The Kriging model was updated using additional samples in the wind-tunnel test, improving its accuracy around the optimum region. Additional samples can also indicate the optimum installation points; more additional sample points improved the maximum lift coefficient, suggesting that the present method using GA could be applied to experimental problems drastically reducing the design cost. Design knowledge was improved by visualizing the Kriging model based on present sampling.
150 Trans. Japan Soc. Aero. Space Sci. Vol. 51, No. 173 One sample that achieved the best performance was selected and its lift curve was compared to that without the nacelle chine. The result showed that the nacelle chine installed at the optimum point increases maximum lift and lift performance at landing was unaffected. We confirmed that the nacelle chine selected using the present method works well when separation occurs due to the nacelle pylon. References Interaction 1) van Dam, C. P.: The Aerodynamic Design of Multi-element High-lift Systems for Transport Airplanes, Prog. Aerospace Sci., 38 (2002), pp. 101 144. 2) http://www.boeing.com/news/frontiers/archive/2003/october/i tt.html, last access on June 1, 2007. 3) Yokokawa, Y., Murayama, M., Ito, T. and Yamamoto, K.: Experiment and CFD of a High-Lift Configuration Civil Transport Aircraft Model, AIAA-2005-2864, 2005. 4) Jeong, S., Murayama, M. and Yamamoto, K.: Efficient Optimization Design Method Using Kriging Model, J. Aircraft, 42 (2005), pp. 413 420. 5) Kanazaki, M., Jeong, S. and Murayama, M.: High-Lift System Optimization Based on Kriging Model Using High Fidelity Flow Solver, T. Jpn. Soc. Aeronaut. Space Sci., 49 (2006), pp. 169 174. 6) Kanazaki, M., Obayashi, S. and Nakahashi, K.: Exhaust Manifold Design with Tapered Ipes Using Divided Range MOGA, Engineering Optimization, Special Issue, 36, 2 (2004), pp. 149 164. 7) Bonaiuti, D. and Pediroda, V.: Aerodynamic Optimization of an Industrial Centrifugal Compressor Impeller Using Genetic Algorithm, Proc. Eurogen 2001, Evolutionary Methods for Design, Optimization and Control, 2002, pp. 467 472. 8) Quagliarella, D. and Cioppa, A. D.: Genetic Algorithms Applied to the Aerodynamic Design of Transonic Airfoils, AIAA Paper 94-1896-CP, 1994. 9) Obayashi, S.: Multi-objective Design Exploration Using Efficient Global Optimization, Proceeding of European Conference on Computational Fluid Dynamics, 2006, (CD-ROM). 10) Tanese, R.: Distributed Genetic Algorithms, Proceedings of 3rd ICGA, 1989, pp. 434 439. Fig. 11. Comparison of velocity magnitude contours measured by PIV: without chine, with chine installed at optimum point.