رابط بهینه سازی شبیه سازی شی گرا

Progress in the field of structural optimization naturally leads to an increasing number of structural models and optimization algorithms that need to be considered for design. Software architecture is of central importance in the ability to account for the complex links tying new structural models and optimizers. An object-oriented programming pattern for interfacing simulation and optimization codes is described in this article. The concepts of optimization variable, criteria, optimizers and simulation environment are the building blocks of the pattern. The resulting interface is logical, flexible and extensive. It encompasses constrained single or multiple objective formulations with continuous, discrete or mixed design variables. Applications are given for composite laminate design.

As research in the field of structural modelling and optimization progresses, design needs to account for an increasing number of interdependent models and allow diverse problem formulations. The architecture of the computer program that implements structural models and optimization algorithms has therefore become crucial in dealing with such increasing complexity. Deliberate programming is a stepping stone for effectively capitalizing knowledge so that creativity in design is not limited by technical book-keeping. This article describes a central part of any structural optimization program, the interface between simulation and optimization sub-programs. The interface is presented as an object-oriented programming pattern. Unlike sequential programming languages (FORTRAN, C, BASIC, PASCAL,…), object-oriented languages like C++ [1] and [2] provide ways to express the solutions to a problem directly and concisely, by creating new relevant types and their functionalities (objects). Object-oriented programming patterns [3] and [4] go one step further: they describe how different objects work together to fulfill a task. In the field of mechanics, object-oriented patterns have mainly covered finite element programming [5], [6], [7] and [8]. The need for object-oriented analysis in advanced optimization strategies has been clearly stated and solutions proposed in [9] and [10]. The text starts with a review of simulation–optimization interfaces. Next, a pattern for an internal interface is proposed, which consists of assigning responsibilities to new programming types related to optimization variables, criteria, algorithm and simulation environment. The resulting program is versatile as it allows constrained single or multiple objective formulations with continuous, discrete or mixed design variables. Applications are given for composite laminate design. A coupled process-structure problem and a coupled material selection-structure problem are solved.

When manufacturing composites laminates, a large numbers of parameters can be controlled that have an influence on the process (e.g., injection pressure, injection gate location, compaction rate), or on the final structure, or even on both process and structure (fibers arrangements, number of plies, choice of materials). In addition, numerous manufacturing and design criteria (e.g., injection maximum pressure or time, stiffness, failure) need to be considered when choosing these parameters. The inherent complexity of composite design make it an important example where the architecture of the simulation–optimization conditions the final design. Two examples have been given to illustrate the point: a process-structure and a material selection-structure coupled optimization. The simulation–optimization interface presented in the text is object-oriented, internal, and is based on four natural classes, optimization variables, criteria, optimizers and simulation environment. The classes are assembled through systematic use of object composition and the abstract factory pattern. It is likely that well-funded interface patterns will become increasingly necessary as more practical optimizers, using all information available about the simulation (approximations, sensitivities, numerical cost of sub-simulations) come into use