This issue of AIEDAM is the second in a series of three “mini” special issues on Evolutionary Design by computers. The papers continue the theme that began in Vol. 13, No. 3, 1999, of using Evolutionary Computation for design problems. The first paper by Eby, Averill, Punch and Goodman provides an excellent overview of the most recent work at Michigan State University on this subject. They describe their work on the optimization of flywheels by an injection island genetic algorithm, and show the importance of minimizing the computation time devoted to evaluation for such real-world applications.

This paper presents an approach to optimal design of elastic flywheels using an Injection Island Genetic Algorithm (iiGA), summarizing a sequence of results reported in earlier publications. An iiGA in combination with a structural finite element code is used to search for shape variations and material placement to optimize the Specific Energy Density (SED, rotational energy per unit weight) of elastic flywheels while controlling the failure angular velocity. iiGAs seek solutions simultaneously at different levels of refinement of the problem representation (and correspondingly different definitions of the fitness function) in separate subpopulations (islands). Solutions are sought first at low levels of refinement with an axi-symmetric plane stress finite element code for high-speed exploration of the coarse design space. Next, individuals are injected into populations with a higher level of resolution that use an axi-symmetric three-dimensional finite element code to “fine-tune” the structures. A greatly simplified design space (containing two million possible solutions) was enumerated for comparison with various approaches that include: simple GAs, threshold accepting (TA), iiGAs and hybrid iiGAs. For all approaches compared for this simplified problem, all variations of the iiGA were found to be the most efficient. This paper will summarize results obtained studying a constrained optimization problem with a huge design space approached with parallel GAs that had various topological structures and several different types of iiGA, to compare efficiency. For this problem, all variations of the iiGA were found to be extremely efficient in terms of computational time required to final solution of similar fitness when compared to the parallel GAs.

This paper evaluates an evolution strategy to tune conventional proportional plus integral plus derivative (PID) and gain scheduling PID control algorithms. The approach deals with the utilization of an evolution strategy with learning acceleration by derandomized mutative step-size control using accumulated information. This technique is useful to obtain the following characteristics: (1) freedom of choice of a performance index, (2) increase of the convergence speed of evolution strategies to get a local minimum to determine controller design parameters, and (3) flexibility and robustness in the automatic design of controllers. Performance analysis and experimental results are carried out using a laboratory scale nonlinear process fan and plate. The practical prototype contains features such as nonminimum phase, dead time, resonant, and turbulent disturbance behavior that motivate the utilization of intelligent control techniques.

An intelligent hybrid approach has been developed to integrate various stages in total design, including formulation of product design specifications, conceptual design, detail design, and manufacture. The integration is achieved by blending multiple artificial intelligence (AI) techniques and CAD/CAE/CAM into a single environment. It has been applied into power transmission system design. In addition to knowledge-based systems and artificial neural networks, another AI technique, genetic algorithms (GAs), are involved in the approach. The GA is used to conduct optimization tasks: (1) searching the best combination of design parameters to obtain optimum design of gears, and (2) optimization of the architecture of the artificial neural networks used in the hybrid system. In this paper, after a brief overview of the intelligent hybrid system, the GA applications are described in detail.

An Artificial Neural Network (ANN) model has been developed for analyzing traffic in an inland waterway network. The main purpose of this paper is to determine how well such a relatively fast model for analyzing a queuing network could substitute for far more expensive simulation. Its substitutability for simulation is judged by relative discrepancies in predicting tow delays between the ANN and simulation models. This model is developed by integrating five distinct ANN submodels that predict tow headway variances at (1) merge points, (2) branching (i.e., diverging) points, (3) lock exits, and (4) link outflow points (e.g., at ports, junctions, or lock entrances), plus (5) tow queuing delays at locks. Preliminary results are shown for those five submodels and for the integrated network analysis model. Eventually, such a network analyzer should be useful for designing, selecting, sequencing, and scheduling lock improvement projects, for controlling lock operations, for system maintenance planning, and for other applications where many combinations of network characteristics must be evaluated. More generally, this method of decomposing complex queuing networks into elements that can be analyzed with ANNs and then recombined provides a promising approach for analyzing other queuing networks (e.g., in transportation, communication, computing, and production systems).

In the course of data modelling, many models could be created. Much work has been done on formulating guidelines for model selection. However, by and large, these guidelines are conservative or too specific. Instead of using general guidelines, models could be selected for a particular task based on statistical tests. When selecting one model, others are discarded. Instead of losing potential sources of information, models could be combined to yield better performance. We review the basics of model selection and combination and discuss their differences. Two examples of opportunistic and principled combinations are presented. The first demonstrates that mediocre quality models could be combined to yield significantly better performance. The latter is the main contribution of the paper; it describes and illustrates a novel heuristic approach called the SG(k-NN) ensemble for the generation of good-quality and diverse models that can even improve excellent quality models.

This paper describes a generic approach to guiding designers when making decisions during the early stages of design. The objective of the research is to enable designers to foresee unintended life-cycle consequences during mechanical component design. Engineering design is a process of evolving solutions to a design problem through the commitment of decisions. As a designer commits a new design decision, a more concrete design solution is generated. Decisions made can have intended and unintended consequences on the performance of the life phase activities that follow, such as manufacturing, assembly, and disposal. Many existing tools only consider the impact of the design solution on later life-cycle phases when the solution is almost complete. This makes changes expensive and difficult. This paper presents a novel approach to how consequences encountered in down stream life-cycle phases can be brought to the designer's attention early in generation of component form. For this purpose, a knowledge model has been derived from a phenomena model. The phenomena model describes how life-cycle consequences are generated during component synthesis. An insight into the representation of the resultant knowledge model is discussed through examples. The implementation of a prototype Knowledge Intensive CAD tool, entitled FORESEE, aimed at supporting life-oriented, feature-based component synthesis and exploration, is also described. The results of the evaluation of FORESEE with a range of designers show that by using the system designers are motivated to explore alternative design solutions and are able to make more informed design decisions. This highlights that the knowledge structure provides a base for extending feature-based component design to a ‘Design Synthesis for Multi-X’ approach.