外文翻譯---用多目標進化算法中的神經(jīng)加速度方法設計一種摩托車車架（英文）

1、J Heuristics (2009) 15: 177–196 DOI 10.1007/s10732-007-9069-4Design of a motorcycle frame using neuroacceleration strategies in MOEAsJorge E. Rodríguez · Andrés L. Medaglia · Carlos A. Coello CoelloRe

2、ceived: 12 February 2007 / Revised: 16 November 2007 / Accepted: 21 December 2007 / Published online: 16 January 2008 © Springer Science+Business Media, LLC 2008Abstract Designing a low-budget lightweight motorcycle

3、 frame with superior dy- namic and mechanical properties is a complex engineering problem. This complexity is due in part to the presence of multiple design objectives—mass, structural stress and rigidity—, the high comp

4、utational cost of the finite element (FE) simulations used to evaluate the objectives, and the nature of the design variables in the frame’s geometry (discrete and continuous). Therefore, this paper presents a neuroaccel

5、era- tion strategy for multiobjective evolutionary algorithms (MOEAs) based on the com- bined use of real (FE simulations) and approximate fitness function evaluations. The proposed approach accelerates convergence to th

6、e Pareto optimal front (POF) com- prised of nondominated frame designs. The proposed MOEA uses a mixed geno- type to encode discrete and continuous design variables, and a set of genetic oper- ators applied according to

7、the type of variable. The results show that the proposed neuro-accelerated MOEAs, NN-NSGA II and NN-MicroGA, improve upon the per- formance of their original counterparts, NSGA II and MicroGA. Thus, this neuroac- celerat

8、ion strategy is shown to be effective and probably applicable to other FE-based engineering design problems.Keywords Multiobjective evolutionary algorithms · Finite element analysis · Neural networks · Mot

9、orcycle · Engineering design · Multiobjective optimizationJ.E. Rodríguez · A.L. Medaglia (? )Centro de Optimización y Probabilidad Aplicada, Departamento de Ingeniería Industrial, Universida

10、d de los Andes, Carrera 1 N. 18A 10, Bogota, Colombia e-mail: amedagli@uniandes.edu.coJ.E. Rodríguez e-mail: je.rodriguez58@egresados.uniandes.edu.coC.A. Coello Coello CINVESTAV-IPN, Departamento de Computación

11、, Ave. Instituto Politécnico Nacional No. 2508, Col. San Pedro Zacatenco, México, D.F., 07300, Mexico e-mail: ccoello@cs.cinvestav.mxDesign of a motorcycle frame using neuroacceleration strategies 179MOEA using

12、 a neural network (NN). Third, the presence of both discrete and contin- uous design variables poses a challenge to the representation of the solution and the genetic operators in the MOEA. Consequently, the proposed alg

13、orithm incorporates a mixed genotype able to store discrete and continuous variables, and a set of ge- netic operators which operate accordingly to the type of variable. Fourth, it expands the object-oriented framework J

14、GA (Medaglia and Gutiérrez 2006) and the MO-JGA (Medaglia et al. 2006) to rapidly implement the proposed algorithms and the alter- native state-of-the-art MOEAs. Lastly, the results of the neuroaccelerated MOEAs are

15、 compared to those produced by their classical counterparts, NSGA II (Deb et al. 2002) and MicroGA (Coello and Toscano 2005), and by the widely used aggregative approach, whose main advantage is its ease of coding and co

16、mputational efficiency (Rodríguez et al. 2005).The remainder of the paper is organized as follows. Section 1 explains the motor-cycle frame design problem and the FE simulation. Section 2 outlines the genotype and g

17、enetic operators used in the MOEAs. Section 3 describes the neuroaccelerated MOEA. Section 4 shows the computational results and the approximate Pareto fron- tiers. Finally, Sect. 5 summarizes the work and suggests futur

18、e extensions of the pro- posed model.1 The motorcycle frame design problemThis work is part of a project aimed to design and construct a low-cost motorcycle (Calderón 2004; Rodríguez et al. 2005). Calderón

19、 (2004) obtained the first functional design of the motorcycle frame using the conventional design approach of manually fine tuning the design variables and running iterative FE simulations. However, this manual approach

20、 can be automated by programmatically simulating the frame’s per- formance using a FE-based software as a black box coupled with an optimization algorithm. Developing such an automated design procedure for a light weight

21、 frame with superior dynamic and mechanical properties requires identification of the design variables and objectives through in-depth study of the motorcycle’s components and their interactions. Indeed, some key attribu

22、tes that impact the motorcycle system’s performance (i.e., stability, maneuverability, and ergonomicity) must be defined prior to decisions about which design variables should be included in the optimization process. The

23、se fixed attributes include the wheelbase, 1,500 mm; the seat height, 640 mm; the rake angle, 36?; and the frame material, carbon steel. For a valuable in- troduction to the dynamic performance of a motorcycle the reader

24、 is referred to Foale (2002).To keep the flexibility of the frame’s geometry while providing a wide range offeasible designs, the design variables domains follow the guidelines in Calderón (2004). A total of 22 para

25、meters are identified and classified, including the contin- uous variables—angles, fillets, lengths, and relative positions of the various frame parts—and the frame parts themselves, which are treated as discrete variabl

26、es. For these latter, given the prohibitive cost of custom-built geometries, this research uses standardized and commercially available tubes and plates (Avallone and Baumeister 1997). Thus, eight tube specifications are