Fabiano Luis de Sousa

New Stochastic Algorithm for Design Optimization

A new stochastic algorithm for design optimization is introduced. Called generalized extremal optimization (GEO), it is intended to be used in complex optimization problems where traditional gradient-based methods may become inefficient, such as when applied to a nonconvex or disjoint design space, or when there are different kinds of design variables in it. The algorithm is easy to implement, does not make use of derivatives, and can be applied to unconstrained or constrained problems, and nonconvex or disjoint design spaces, in the presence of any combination of continuous, discrete, or integer variables. It is a global search metaheuristic, as are genetic algorithms (GAs) and simulated annealing (SA), but with the a priori advantage of having only one free parameter to adjust. The algorithm is presented in two implementations and its performance is assessed on a set of test functions. A simple application to the design of a glider airfoil is also presented. It is shown that the GEO algorithm is competitive in performance with the GA and SA and is an attractive tool to be used on applications in the aerospace field.

Heat Pipe Design Through Generalized Extremal Optimization

In this paper, an application of the Generalized Extremal Optimization (GEO) algorithm to the optimization of a heat pipe (HP) for a space application is presented. The GEO algorithm is a generalization of the Extremal Optimization (EO) algorithm, devised to be applied readily to a broad class of design optimization problems regardless of the design space complexity it would face. It is easy to implement, does not make use of derivatives, and can be applied to either unconstrained or constrained problems with continuous, discrete, or integer variables. The GEO algorithm has been tested in a series of test functions and shows to be competitive to other stochastic algorithms, such as the Genetic Algorithm. In this work, it is applied to the problem of minimizing the mass of an HP as a function of a desirable heat transport capability and a given temperature on the condenser. The optimal solutions were obtained for different heat loads, heat sink temperatures, and three working fluids: ammonia, methanol, and ethanol. The present design application highlights the GEO features of being easily implemented and efficient on tackling optimization problems when the objective function presents design variables with strong nonlinear interactions and is subject to multiple constraints.

Generating Feasible Test Paths from an Executable Model Using a Multi-objective Approach

Search-based testing techniques using meta-heuristics, like evolutionary algorithms, has been largely used for test data generation, but most approaches were proposed for white-box testing. In this paper we present an evolutionary approach for test sequence generation from a behavior model, in particular, Extended Finite State Machine. An open problem is the production of infeasible paths, as these should be detected and discarded manually. To circumvent this problem, we use an executable model to obtain feasible paths dynamically. An evolutionary algorithm is used to search for solutions that cover a given test purpose, which is a transition of interest. The target transition is used as a criterion to get slicing information, in this way, helping to identify the parts of the model that affect the test purpose. We also present a multi-objective search: the test purpose coverage and the sequence size minimization, as longer sequences require more effort to be executed.

Spacecraft thermal design with the Generalized Extremal Optimization Algorithm

This article describes an application of the Generalized Extremal Optimization (GEO) algorithm to the inverse design of a spacecraft thermal control system. GEO is a recently proposed global search meta-heuristic (Sousa, F.L. and Ramos, F.M., 2002, Function optimization using extremal dynamics. In: Proceedings of the 4th International Conference on Inverse Problems in Engineering (cd-rom), Rio de Janeiro, Brazil.; Sousa, F.L., Ramos, F.M., Paglione, P. and Girardi, R.M., 2003, New stochastic algorithm for design optimization. AIAA Journal, 41(9), 1808–1818.; Sousa, F.L., Ramos, F.M., Galski, R.L. and Muraoka, I., 2005, Chapter III. In: L.N. De Castro and F.J. Von Zuben (Eds) Generalized Extremal Optimization: A New Meta-heuristic Inspired by a Model of Natural Evolution, Accepted for publication in Recent Developments in Biologically Inspired Computing (Hershey, PA: Idea Group Inc.).) based on a model of natural evolution (Bak, P. and Sneppen, K., 1993, Punctuated equilibrium and criticality in a simple model of evolution. Physical Review Letters, 71(24), 4083–4086), and specially devised to be used in complex optimization problems (Sousa, F.L., Vlassov, V. and Ramos, F.M., 2002, Heat pipe design through generalized extremal optimization. In: Proceedings of the IX Brazilian Congress of Engineering and Thermal Sciences – ENCIT 2002, Caxambu, MG, Brazil.). GEO is easy to implement, has only one free parameter to adjust, does not make use of derivatives and can be applied to constrained or unconstrained problems, and to non-convex or even disjoint design spaces with any combination of continuous, discrete, or integer variables. The application reported here concerns the optimum design of a simplified configuration of the Brazilian Multi-mission Platform (in Portuguese, Plataforma Multi-missão, PMM) thermal control subsystem, comprising five radiators and one battery heater. The PMM is a multi-purpose space platform to be used in different types of missions such as Earth observation, scientific, or meteorological data collecting. The design procedure is tackled as a multiobjective optimization problem, considering two critical cases, operational hot case (HC) and cold case (CC). The results indicate the existence of non-intuitive, new and more efficient design solutions.

MOST: A Multi-objective Search-Based Testing from EFSM

This paper introduces a multi-objective evolutionary approach to test case generation from extended finite state machines (EFSM), named MOST. Testing from an (E)FSM generally involves executing various transition paths, until a given coverage criterion (e.g. cover all transitions) is met. As traditional test generation methods from FSM only consider the control aspects, they can produce many infeasible paths when applied to EFSMs, due to conflicts in guard conditions along a path. In order to avoid the infeasible path generation, we propose an approach that obtains feasible paths dynamically, instead of performing static reach ability analysis as usual for FSM-based methods. Previous works have treated EFSM test case generation as a mono-objective optimization problem. Our approach takes two objectives into account that are the coverage criterion and the solution length. In this way, it is not necessary to establish in advance the test case size as earlier approaches. MOST constructs a Pareto set approximation, i.e., a group of optimal solutions, which allows the test team to select the solutions that represent a good trade-off between both objectives. The paper shows empirical studies to illustrate the benefits of the approach and comparing the results with the ones obtained in a related work.

Method of glass selection for color correction in optical system design

A method of glass selection for the design of optical systems with reduced chromatic aberration is presented. This method is based on the unification of two previously published methods adding new contributions and using a multi-objective approach. This new method makes it possible to select sets of compatible glasses suitable for the design of super-apochromatic optical systems. As an example, we present the selection of compatible glasses and the effective designs for all-refractive optical systems corrected in five spectral bands, with central wavelengths going from 485 nm to 1600 nm.

Inverse problems in space science and technology

Solutions for inverse problems appearing in space applications and space technology are described. The inverse problem is formulated as a nonlinear optimization problem. Usually some additional information must be added from our previous knowledge about the physical phenomenon. In general this a priori information means smoothness, in other words, regularized solutions are searched for. The methodology is applied to geophysics (magneto-telluric inversion), meteorology (temperature retrieval from satellite data), and oceanography (inverse hydrologic optics), as examples of space applications. The scheme is also employed for solving an inverse problem emerging from technology: the inverse design of a space radiator.

Generalized extremal optimization: an attractive alternative for test data generation

Software testing is an important activity of the software development process, and its automation (specially of test data generation) has been a burgeoning interest of many researchers [1]. It has recently been shown that evolutionary algorithms, such as the Genetic Algorithms (GAs), are valuable tools for test data generation. This work assesses the performance of a recently proposed evolutionary algorithm, the Generalized Extremal Optimization (GEO) [2], on test data generation for programs that have paths with loops. GEO’s main advantage in comparison to other stochastic algorithms is that it has only one free parameter to adjust, which eases the process to set it to give its best performance in a given application. Test data generation consists of generating inputs for the SUT in order to evaluate its internal (white-box testing) or external (black-box testing) behavior. Typically, path testing consists of two steps: (i) select a finite set of paths to be exercised; (ii) generate test data to execute the set of paths. For the first step a criterion is necessary, since testing all execution paths in a program is generally impossible due to the existence of infeasible paths and loops. Here our concern is on step (ii): given a set of paths, how do you generate test data to exercise them?

Application of a GEO + SA hybrid optimization algorithm to the solution of an inverse radiative transfer problem

In a former study (F.L. de Sousa, F.M. Ramos, F.J.C.P. Soeiro, and A.J. Silva Neto, Application of the generalized extremal optimization algorithm to an inverse radiative transfer problem, Inverse Probl. Sci. Eng. 15 (2007), pp. 699–714), a new evolutionary optimization metaheuristic–the generalized extremal optimization (GEO) algorithm (F.L. de Sousa, F.M. Ramos, P.Paglione, and R.M. Girardi, A new stochastic algorithm for design optimization, AIAA J. 41 (2003), pp. 1808–1818)–was applied to the solution of an inverse problem of radiative properties estimation. A comparison with two other stochastic methods; simulated annealing (SA) and genetic algorithms (GA), was also performed, demonstrating GEOs competitiveness for that problem. In the present article, a recently developed hybrid version of GEO and SA (R.L. Galski, Development of improved, hybrid, parallel, and multiobjective versions of the generalized extremal optimization method and its application to the design of spatial systems, D.Sc. Thesis, Instituto Nacional de Pequisas Espaciais, Brazil, 2006, p. 279. INPE-14795-TDI/1238 (in Portuguese)) is applied to the same radiative transfer problem and the results obtained are compared with those from the previous study. The present approach was already foreseen (e.g. in F.L. de Sousa, F.M. Ramos, F.J.C.P. Soeiro, and A.J. Silva Neto, Application of the generalized extremal optimization algorithm to an inverse radiative transfer problem, Inverse Probl. Sci. Eng. 15 (2007), pp. 699–714) as a technique that could significantly improve the performance of GEO for this problem. The idea is to make use of a scheduling for GEOs free parameter γ in a similar way to the cooling rate of SA. The main objective of this approach is to combine the good exploration properties of GEO during the early stages of the search with the good convergence properties of SA at the end of the search.

A multi-objective evolutionary algorithm to obtain test cases with variable lengths

In this paper a new multi-objective implementation of the generalized extremal optimization (GEO) algorithm, named M-GEOvsl, is presented. It was developed primarily to be used as a test case generator to find transition paths from extended finite state machines (EFSM), taking into account not only the transition to be covered but also the minimization of the test length. M-GEOvsl has the capability to deal with strings whose number of elements vary dynamically, making it possible to generate solutions with different lengths. The steps of the algorithm are described for a general multi-objective problem in which the solution length is an element to be optimized. Experiments were performed to generate test case from EFSM benchmark models using M-GEOvsl and the approach was compared with a related work.