Civil and Structural Engineering Computing: 2001
Chapter 7 T. Arciszewski+ and K.A. De Jong*
+Department of Civil, Environmental, and Infrastructure Engineering, George Mason University, Fairfax, United States of America *Computer Science Department, George Mason University, Fairfax, United States of America Keywords: evolutionary computating, evolutionary design, conceptual design, multi-criteria optimization, integrated design, morphogenic evolution, co-evolutionary design, complex adaptive system, visualization
Among existing computational paradigms, evolutionary computation (EC) is now recognized as particularly appropriate for various traditional and novel computational applications in civil engineering. EC involves the use of evolutionary algorithms (EAs) to solve difficult problems in science and engineering. An EA is an algorithm that captures a Darwinian notion of ``survival of the fittest" in a computationally useful form. In the paper a unified view of EAs is presented including a general characterization of EAs from the engineering perspective using a 7-attribute description. This general characterization is then used to describe three historically important EAs: Evolution Strategies, Evolutionary Programming, and Genetic Algorithms. These evolutionary algorithms are likely to become a fundamental part of the engineering design process and subsequently may be incorporated into various design support tools. Their true potential in the context of a holistic approach to engineering design is still not completely known, but this issue is discussed in some detail in the paper from the perspective of integrated design. In the paper integrated design is understood as a design process in which both the conceptual and detailed design processes are fully integrated and conducted by a single design support tool. The increasing complexity of application designs raises some difficult internal EA issues relating to how best to represent and evolve complex designs. If we look to nature for inspiration here, we observe evolution manipulating the genetic plans for complex objects rather than the objects themselves through the process called morphogenesis. As a consequence, the authors decided from the beginning of their research, that the next generation of evolutionary design tools should support an integrated ``morphogenic" design process. In this case, an EA is manipulating conceptual designs and the fitness evaluation of a conceptual design is based on producing a detailed design from the conceptual design and obtaining the most accurate analytical data available through the use of state of the art computer packages for the analysis, design and optimization of structural systems. Thus, all feasible design concepts are ``transformed" into detailed designs and complete designs are produced during the evolutionary process. In this way, the designer receives only final complete designs produced by such an integrated design tool. In order to explore this idea of a morphogenic design process, an experimental design and research tool has been developed by the authors at GMU, called ``Inventor 2000". It is intended for design experiments in the area of wind bracings in steel skeleton structures of tall buildings allowing a complete design for buildings of various dimensions and height, within range 16 – 36 stories. Inventor 2000 produces both the design concepts and the detailed designs. It is briefly described in the paper. There is an increasing interest in understanding and using co-evolutionary systems to solve difficult problems. Such systems have multiple populations of different ``species" that affect each other's fitness and can be used in a number of promising ways for evolutionary design. The paper provides a state of the art review in this area and proposes a unified classification of co-evolution in design using a 7-attribute description. Evolutionary algorithms are themselves complex, non-linear systems that need to be understood in sufficient detail that they can be properly used to effectively explore design spaces. Therefore, the paper briefly describes three research projects at George Mason University dealing with the analysis of evolutionary computation and its results. In the first case, a complex adaptive system approach is used to study dynamics of evolutionary computation. In the two other projects, various forms of visualization of results are investigated and visualization tools are built. The paper provides various conclusions and suggestions for further research. Among the most important conclusions are those about the evolving nature of computations in civil engineering and about the growing importance of evolutionary computation, particularly in the area of design. In this changing situation, an educational effort is necessary to introduce fundamentals of evolutionary computation within the civil engineering community, including computational principles, models, methods, and tools. Also, the authors postulate more research on both morphogenic and co-evolutionary design. Finally, the authors believe that research should be also concentrated on the methods and tools for the analysis of results of complex evolutionary processes in design, including the use of a complex adaptive system approach and various visualization methods and tools.
| ||||||||||
