Kazuhisa Chiba

Data Mining for Aerodynamic Design Space

Analysis of variance (ANOVA) and self-organizing map (SOM) were applied to data mining for aerodynamic design space. These methods make it possible to identify the effect of each design variable on objective functions. ANOVA shows the information quantitatively, while SOM shows it qualitatively. Furthermore, ANOVA can show the effects of interaction between design variables on objective functions and SOM can visualize the trade-offs among objective functions. This information will be helpful for designers to determine the final design from non-dominated solutions of multi-objective problems. These methods were applied to two design results: a fly-back booster in reusable launch vehicle design, which has 4 objective functions and 71 design variables, and a transonic airfoil design performed with the adaptive search region method.

Multi-Objective Design Exploration for Aerodynamic Configurations

A new approach, Multi-Objective Design Exploration (MODE), is presented to address Multidisciplinary Design Optimization problems. MODE reveals the structure of the design space from the trade-off information and visualizes it as a panorama for Decision Maker. The present form of MODE consists of Kriging Model, Adaptive Range Multi Objective Genetic Algorithms, Analysis of Variance and Self-Organizing Map. The main emphasis of this approach is visual data mining. Two data mining examples using high fidelity simulation codes are presented: four-objective aerodynamic optimization for the fly-back booster and Multidisciplinary Design Optimization problem for a regional-jet wing. The first example confirms that two different data mining techniques produce consistent results. The second example illustrates the importance of the present approach because design knowledge can produce a better design even from the brief exploration of the design space.

Multidisciplinary Design Optimization and Data Mining for Transonic Regional-Jet Wing

A large-scale, real-world application of evolutionary multi-objective optimization is reported. The multidisciplinary design optimization among aerodynamics, structures, and aeroelasticity of the wing of a transonic regional-jet aircraft was performed using high-fidelity evaluation models. Euler and Navier-Stokes solvers were employed for aerodynamic evaluation. The commercial software NASTRAN was coupled with a computational fluid dynamics solver for the structural and aeroelastic evaluations. An adaptive range multi-objective genetic algorithm was employed as an optimizer. The objective functions were minimizations of block fuel and maximum takeoff weight in addition to drag divergence between transonic and subsonic flight conditions. As a result, nine nondominated solutions were generated and used for tradeoff analysis among three objectives. Moreover, all solutions evaluated during the evolution were analyzed using a self-organizing map as a data mining technique to extract key features of the design space. One of the key features found by data mining was the nongull wing geometry, although the present multidisciplinary design optimization results showed the inverted gull wings as nondominated solutions. When this knowledge was applied to one optimum solution, the resulting design was found to have better performance and to achieve 3.6% improvement in the block fuel compared to the original geometry designed in the conventional manner.

High-Fidelity Multidisciplinary Design Optimization of Aerostructural Wing Shape for Regional Jet

A large-scale, real-world application of Evolutionary Multi-Objective Optimization is reported. The Multidisciplinary Design Optimization among aerodynamics, structures, and aeroelasticity of the wing of a transonic regional jet aircraft was performed using highfidelity evaluation models. Euler and Navier-Stokes solvers were employed for aerodynamic evaluation. The commercial software NASTRAN was coupled with a Computational Fluid Dynamics solver for the structural and aeroelastic evaluations. Adaptive Range MultiObjective Genetic Algorithm was employed as an optimizer. The objective functions were minimizations of block fuel and maximum takeoff weight in addition to drag divergence between transonic and subsonic flight conditions. As a result, nine non-dominated solutions were generated and used for tradeoff analysis among three objectives. Moreover, all solutions evaluated during the evolution were analyzed using a Self-Organizing Map as a Data Mining technique to extract key features of the design space. One of the key features found by Data Mining was the non-gull wing geometry, although the present MDO results showed the reverse-gull wings as non-dominated solutions. When this knowledge was applied to one optimum solution, the resulting design was found to have better performance and to achieve 3.6 percent improvement in the block fuel compared to the original geometry designed in the conventional manner.

Design-Informatics Approach for Intimate Configuration of Silent Supersonic Technology Demonstrator

The intimate configuration of the silent supersonic technology demonstrator has been designed using the design-informatics approach. As a first step, multidisciplinary design optimization with multi-objectives has been performed for the wing shape of a silent supersonic technology demonstrator among aerodynamics, structures, aeroelasticity, and boom noise. Aerodynamic evaluation was carried out by solving Euler equations on computational fluid dynamics, and composite structural weight evaluation was performed by using MSC. NASTRAN for strength and flutter requirements on computational structural dynamics. The intensity of sonic boom was evaluated by a modified linear theory. The optimization problem had four objective functions as the minimizations of the pressure drag and the boom intensity at supersonic condition, and the composite structural weight. The intimate configuration defined by 50 design variables was optimized on particle swarm optimization and genetic algorithm hybrid method. In the structural evaluation, the combination optimization of stacking sequences of laminated composites was performed for inboard and outboard wings with strength and flutter requirements. Consequently, 37 non-dominated solutions were obtained. As a second step, data mining has been performed to obtain the design knowledge for deciding a compromise solution. The data mining revealed the knowledge in the design space, such as the tradeoff information among the objective functions, and the correlations between objective functions and design variables. A compromise solution was successfully determined by using the obtained design knowledge. Design-informatics approach is essential for an efficient design process.

Knowledge Discovery for Flyback-Booster Aerodynamic Wing Using Data Mining

Data mining has been performed on the results of an aerodynamic design optimization of a two-stage-to-orbit reusable launch vehicle flyback-booster wing. Three data mining techniques were compared, including selforganizing map, functional analysis of variance, and the rough set theory. The optimization problem had four aerodynamic objective functions and 71 wing shape design variables. The hypothetical design database resulting from the optimization contained a total of 302 solutions which included 102 nondominated solutions. Consequently, the acquired knowledge of the design space consisted of general design characteristics, correlation between objective functions, and the effects of these design variables on the objective functions, for both nondominated as well as all solutions. The comparison also revealed the similarities anddifferences among the three datamining techniques used in this study. Even though all three techniques discovered detailed design knowledge and the results produced by the combination of all three methods compensated disadvantages of each method when applied individually, it was discovered that the self-organizing map produced the overall best results. Moreover, this study has also shown that the knowledge acquired from both nondominated solutions and from all solutions found was consistent despite the differences between the design spaces. Furthermore, it was shown that datamining is essential for visualizing results of an evolutionary multi-objective optimization problem and extracting useful design knowledge from these results.

Knowledge Discovery in Aerodynamic Design Space for Flyback-Booster Wing Using Data Mining

The data mining has been performed for the aerodynamic design optimization result of two-stage-to-orbit reusable launch vehicle flyback booster wing. Three data mining techniques were used such as self-organizing map, functional analysis of variance, and rough set theory. The optimization problem had four aerodynamic objective functions and 71 design variables regarding wing shape. The optimization obtained the result as the hypothetical design database with 302 all solutions including the 102 non-dominated solutions. Consequently, the knowledge in the design space was acquired regarding the correlation between objective functions, and the influence of the design variables to the objective function, for non-dominated and all evaluated solutions, respectively. The features of three data mining techniques were revealed. Although the combination among three techniques discovered detailed design knowledge, self-organizing map was especially a key technique for knowledge discovery. Moreover, design knowledge from all solutions conserved the information from non-dominated solutions. Data mining was essential to solve multi-objective optimization problem.

Evolutionary hybrid computation in view of design information by data mining

Design Informatics has three points of view. First point is the efficient exploration in design space using evolutionary computation. Second point is the structurization and visualization of design space using data mining. Third point is the application to practical problems. In the present study, the influence of the seven pure and hybrid optimizers for design information has been investigated in order to explain the selection manner of optimizer for data mining. A single-stage hybrid rocket design problem is picked up as the present design object. As a result, mining result depends on not the number of generation (convergence) but the optimizers (diversity). Consequently, the optimizer with diversity performance should be selected in order to obtain global design information in the design space. Therefore, the diversity performance has also been explained for the seven optimization methods by using three standard mathematical test problems with/without noise. The result indicates that the hybrid method between the differential evolution and the genetic algorithm is beneficial performance for efficient exploration in the design space under the condition for large-scale design problems within 102 order evolution at most. Moreover, the comparison among eight crossovers indicates that the principal component analysis blended crossover is good selection on the hybrid method between the differential evolution and the genetic algorithm.

High-Fidelity multidisciplinary design optimization of wing shape for regional jet aircraft

A large-scale, real-world application of Evolutionary Multi- Criterion Optimization (EMO) is reported in this paper. The Multidisciplinary Design Optimization among aerodynamics, structures and aeroelasticity for the wing of a transonic regional jet aircraft has been performed using high-.delity models. An Euler/Navier-Stokes (N-S) Computational Fluid Dynamics (CFD) solver is employed for the aerodynamic evaluation. The NASTRAN, a commercial software, is coupled with a CFD solver for the structural and aeroelastic evaluations. Adaptive Range Multi-Objective Genetic Algorithm is employed as an optimizer. The objective functions are minimizations of block fuel and maximum takeo. weight in addition to di.erence in the drag between transonic and subsonic .ight conditions. As a result, nine non-dominated solutions have been generated. They are used for tradeo. analysis among three objectives. One solution is found to have one percent improvement in the block fuel compared to the original geometry designed in the conventional manner. All the solutions evaluated during the evolution are analyzed by Self-Organizing Map to extract key features of the design space.

CFD Visualization of Second Primary Vortex Structure on a 65-Degree Delta Wing

Numerical simulation has been performed corresponding to recent experiment around delta wings with sharp and blunt leading edges at NASA Langrey Research Center, which indicates the second primary vortex and quantitative Reynolds-number effects. Three one-equation turbulence models are examined on the unstructured hybrid mesh and the modified Spalart-Allmaras turbulence model is found most effective to capture a complex vortex structure. The adaptive mesh refinement method at a vortex center is also applied. Visualization of the computational results suggests that the second primary vortex may be a developing share layer merging to an open separation of the primary vortex. Not only the volume-mesh refinement but also the surface-mesh refinement is found important to capture Reynolds-number effects around a delta wing with a blunt leading edge. Introduction D wing has been used for space transport and supersonic transport because of high aerodynamic performance. Those transports utilize leadingedge separation at high angles of attack for take-off and landing. Analyses of the leading-edge separation have been performed by many experiments and computations. Previous numerical works about the leading-edge separation around a delta wing are given by for example, Ekaterinaris and Schiff, and Murayama et al. Recent experiment at NASA Langley Research Center investigated effects of leading-edge bluntness and Reynolds-number difference. In this experiment, the sharp and blunt leading edges are used. The sharp leading edge produces a typical conical vortex structure. Suction peak due to the leading-edge separation occurs almost at the same semispan locations for the entire wing. While the blunt leading edge produces a more complex flow. This leading edge delays ∗Graduate Student, Department of Aeronautics and Space Engineering. Student Member AIAA. †Professor, Institute of Fluid Science. Associate Fellow AIAA. ‡Professor, Department of Aeronautics and Space Engineering. Associate Fellow AIAA. Copyright c