Constraint processing has emerged as an extremely expressive and powerful paradigm in which to model, solve, and reason about many complex problems. Over the past four decades advances in both the fundamental aspects of constraint processing and practical applications of constraints have contributed to making it one of the most promising of artificial intelligence technologies.

Previous applications of genetic programming (GP) have been restricted to searching for algebraic approximations mapping the design parameters (e.g., geometrical parameters) to a single design objective (e.g., weight). In addition, these algebraic expressions tend to be highly complex. By adding a simple extension to the GP technique, a powerful design data analysis tool is developed. This paper significantly extends the analysis capabilities of GP by searching for multiple simple models within a single population by splitting the population into multiple islands according to the design variables used by individual members. Where members from different islands “cooperate,” simple design models can be extracted from this cooperation. This relatively simple extension to GP is shown to have powerful implications to extracting design models that can be readily interpreted and exploited by human designers. The full analysis method, GP heuristics extraction method, is described and illustrated by means of a design case study.

This Special Issue is in an exceptional format. Starting with an invited Keynote by Rob Woodbury and Andrew Burrow, the rest of the issue is taken up by invited papers that are responses to Woodbury and Burrow's paper, written by selected authorities on design. A final short section is devoted to Woodbury and Burrow's reactions to the responses.

Style is a word that people think they understand. Most people recognize artifacts like buildings and clothes as being exemplars of particular styles, and they know words like Rococo and Art Deco as names for styles. They can recognize stylistic similarities not only in one sort of artifact but also across wide ranges of different things, such as buildings, furniture, artworks, clothes, music, and even manners. However, “style” is a slippery notion: the word has been used in a variety of senses since the ancient Greeks first thought about the differences in how people wrote or painted, and it is still used to refer to different things.

Analyses of styles in design have paid little attention to how people see style and how designers use perceptions of style to guide designing. Although formal and computational methods for analyzing styles and generating designs provide impressively parsimonious accounts of what some styles are, they do not address many of the factors that influence how humans understand styles. The subtlety of human style judgments raises challenges for computational approaches to style. This paper differentiates between a range of distinct meanings of “style” and explores how designers and ordinary people learn and apply perceptual similarity classes and style concepts in different situations to interpret and create designed artifacts. A range of psychological evidence indicates that style perception is dependent on knowledge and involves the interaction of perceptual recognition of style features and explanatory inference processes that create a coherent understanding of an object as an exemplar of a style. This article concludes by outlining how formal style analyses can be used in combination with psychological research to develop a fuller understanding of style perception and creative design.

In constraint-based design, components are modeled by variables describing their properties and subject to physical or mechanical constraints. However, some other constraints are difficult to represent, like comfort or user satisfaction. Partially defined constraints can be used to model the incomplete knowledge of a concept or a relation. Instead of only computing with the known part of the constraint, we propose to complete its definition by using machine-learning techniques. Because constraints are actively used during solving for pruning domains, building a classifier for instances is not enough: we need a solver able to reduce variable domains. Our technique is composed of two steps: first we learn a classifier for the constraint's projections and then we transform the classifier into a propagator. We show that our technique not only has good learning performances but also yields a very efficient solver for the learned constraint.

Design space exploration is a long-standing focus in computational design research. Its three main threads are accounts of designer action, development of strategies for amplification of designer action in exploration, and discovery of computational structures to support exploration. Chief among such structures is the design space, which is the network structure of related designs that are visited in an exploration process. There is relatively little research on design spaces to date. This paper sketches a partial account of the structure of both design spaces and research to develop them. It focuses largely on the implications of designers acting as explorers.

The recognition and identification of parts are important processes in modern manufacturing systems. Although machine vision systems have played an important role in these tasks, there are still challenges in performing these tasks in which parts may be in motion and subjected to noise. Using a flexible vibratory bowl feeder system as a test bed to simulate motion of parts subjected to noise, scanned signatures of part features are acquired using fiber optic sensors and a data acquisition system. Because neural networks have been shown to exhibit good pattern recognition capability, ARTMAP, a neural network that learns patterns under supervision, was incorporated into the feeder system. The pattern recognition capability of the feeder system is dependent on a set of parameters that characterized ARTMAP, the sampling rate of the data acquisition system, and the mean speed of the vibrating parts. The parameters that characterized ARTMAP are the size of an input vector, the vigilance, threshold value of the nonlinear noise suppression function, and the learning rate. Through extensive training and testing of the ARTMAP within the feeder system, it was shown that high success rates of recognition of parts features in motion under noisy conditions can be obtained provided these parameters of ARTMAP are appropriately selected.

The paper by Woodbury and Burrow is examined using four criteria: completeness, discrimination, alternative approaches, and combining exploration in different problem domains. Although the paper covers significant aspects of the search space literature it leaves out some relevant aspects of the cognitive approach. It fares much better in terms of discriminating important concepts and alternative approaches to the modeling of the design search space. The structure–function–behavior model is suggested as an analogy for the central parameters of the search space paradigm. The Woodbury and Burrow paper reveals more than what has been accomplished up to now in the design search space area, but its task still remains incomplete.

High speed machinery has played and continues to play a vital role in the manufacture and production of consumer goods. In the design of high speed systems there are two key considerations: power transmission and motion control. Although there is considerable computer-based support for the design of systems to achieve requirements of power transmission, there is only limited support for the design of systems to deliver complex motion control. This is particularly the case where mechanism and linkage systems are considered in order to achieve large displacements and intricate paths involving reentrant and reciprocating components. One explanation for this relative lack of supportive tools is the underlying reasoning and analysis techniques implemented within many commercial and research software environments. To overcome these limitations a constraint-based approach has been employed to provide the fundamental elements of a design environment for mechanisms and machine systems. The design environment provides support for the transition from concept to embodiment stages of the design process and the subsequent stages of detailed design and optimization. In contrast to many research approaches the design environment presented in this paper has been created and developed through close collaboration with industry and through extensive application to real design scenarios. First, the underlying representations and methods are presented. The fundamental elements of the design environment are then described and its capabilities discussed with particular reference to the use of constraints in design.

At its root level, style is actually an esthetic agreement between people. The question is, how can esthetic agreements be modeled and measured in artificial intelligence? This paper offers a formal theory called EVE′ and applies it to a novel test bed of dynamic drawings that combine features of music and sketching. The theory provides mathematical measures of expectations, violations, and explanations, which are argued to be the atomic components of the esthetic experience. The approach employs Bayesian methods to extend information measures proposed in other research. In particular, it is shown that information theory is useful at an entropic level to measure expectations (E) of signals and violations (V) of expectations, but that Bayesian theory is needed at a semantic level to measure explanations (E′) of meaning for the signals. The entropic and semantic measures are then combined in further measures of tension and pleasure at an esthetic level that is actually style.

Language plays at least two roles in design. First, language serves as representations of ideas and concepts through linguistic behaviors that represent the structure of thought during the design process. Second, language also performs actions and creates states of affairs. Based on these two perspectives on language use in design, we apply the computational linguistics tools of latent semantic analysis and lexical chain analysis to characterize how design teams engage in concept formation as the accumulation of knowledge represented by lexicalized concepts. The accumulation is described in a data structure comprised by a set of links between elemental lexicalized concepts. The folding together of these two perspectives on language use in design with the information processing theories of the mind afforded by the computational linguistics tools applied creates a new means to evaluate concept formation in design teams. The method suggests that analysis at a linguistic level can characterize concept formation even where process-oriented critiques were limited in their ability to uncover a formal design method that could explain the phenomenon.

Generative specifications have been used to systematically codify established styles in several design fields including architecture and product design. We examine how designers explore new designs in the early stages of product development as they manipulate and interpret shape representations. A model of exploration is proposed with four types of shape descriptions (contour, decomposition, structure, and design) and the results of the exploration are presented. Generative rules are used to provide consistent stylistic changes first within a given decomposition and second through changing the structure. Style expresses both the analytical order of explanation and the synthetic complexity of exploration. The model of exploration is consistent with observations of design practice. The application of generative design methods demonstrates a logical pattern for early stage design exploration. The model provides the basis for tools to assist designers in exploring families of designs in a style and for following new interpretations that move the exploration from one family to another.

The problem investigated in this research is that engineering design decision making can be complicated and made difficult by highly coupled design parameters and the vast number of design parameters. This complication often hinders the full exploration of a design solution space in order to generate optimal design solution. These hindrances result in inferior or unfit design solutions generated for a given design problem due to a lack of understanding of both the problem and the solution space. This research introduces a computational framework of a new algebraic constraint-based design approach aimed at providing a deeper understanding of the design problem and enabling the designers to gain insights to the dynamic solution space and the problem. This will enable designers to make informed decisions based on the insights derived from parameter relationships extracted. This paper also describes an enhanced understanding of an engineering design process as a constraint centered design. It argues that with more effort and appreciation of the benefits derived from this constraint-based design approach, engineering design can be advanced significantly by first generating a more quantitative product design specification and then using these quantitative statements as the basis for constraint-based rigorous design. The approach has been investigated in the context of whole product life-cycle design and multidisciplinary design, aiming to derive a generic constraint-based design approach that can cope with life-cycle design and different engineering disciplines. A prototype system has been implemented based on a constraint-based system architecture. The paper gives details of the constraint-based design process through illustrating a worked real design example. The successful application of the approach in two highly coupled engineering design problems and the evaluation undertaken by a group of experienced designers show that the approach does provide the designers with insights for better exploration, enabled by the algebraic constraint solver. The approach thus provides a significant step towards fuller scale constraint-based scientific design.

Having no fundamental difficulties with Woodbury and Burrow's article, I explore some implications of their work based on my experience, and that of PhD students advised by me, with developing design support systems. I suggest that such systems need a distinct task layer on top of the computational layer, where the power of a system rests. I also express unease with some details of the article under review, in the hope of contributing to a wider discussion.

In the push for new vehicle designs, the distinctions between vehicle classes are quickly becoming blurred. We use shape grammars to quantify the differences between vehicle classes through the application of class-specific rules and the constraint of rule applications to within parametric ranges determined for each class. This allows for the development of new vehicle forms that clearly fall within a class or purposefully cross over the boundaries of classes and mix rules and ranges to create unique and interesting cross-over vehicles.

In this paper, we present the development and application of a technical feasibility model used in preliminary design to determine whether a set of desired product specifications obtained from marketing is feasible in the engineering domain. This model is developed by integrating the capabilities of a multiobjective design problem, a multicriteria design optimization tool, a Pareto frontier gap analyzer, metamodeling methods, and use of the Pareto frontier as a constraint for feasibility assessment. Although the tools are independent of the domain, their application is illustrated using two examples: a simple three-objective mathematical problem and a five-objective passenger vehicle design problem. The feasibility of the desired product specifications is determined with respect to the problem's Pareto frontier, which is considered to be the necessary constraint to satisfy.

This paper examines the need for explicit representations of the design space, in response to Woodbury and Burrow. Specifically, their proposal for a particular search strategy, by means of which one can reuse past experiences explicitly represented by previously traversed paths, is examined. This is done by exploring issues with respect to design search and representation in general, while relating these to specific issues raised by Woodbury and Burrow. The paper concludes by suggesting that their arguments essentially point to devising an appropriate “programming language” for design.

Style is an ordering principle by which to structure artifacts in a design domain. The application of a visual order entails some explicit grouping property that is both cognitively plausible and contextually dependent. Central to cognitive–contextual notions are the type of representation used in analysis and the flexibility to allow semantic interpretation. We present a model of visual style based on the concept of similarity as a qualitative context-dependent categorization. The two core components of the model are semantic feature extraction and self-organizing maps (SOMs). The model proposes a method of categorizing two-dimensional unannotated design diagrams using both low-level geometric and high-level semantic features that are automatically derived from the pictorial content of the design. The operation of the initial model, called Q-SOM, is then extended to include relevance feedback (Q-SOM:RF). The extended model can be seen as a series of sequential processing stages, in which qualitative encoding and feature extraction are followed by iterative recategorization. Categorization is achieved using an unsupervised SOM, and contextual dependencies are integrated via cluster relevance determined by the observer's feedback. The following stages are presented: initial per feature detection and extraction, selection of feature sets corresponding to different spatial ontologies, unsupervised categorization of design diagrams based on appropriate feature subsets, and integration of design context via relevance feedback. From our experiments we compare different outcomes from consecutive stages of the model. The results show that the model provides a cognitively plausible and context-dependent method for characterizing visual style in design.

The paper presents aspects of designer action that stress cognitive strategies for effective design problem solving, under the headings of exploration and representation. It proposes that links among design moves and shifts between design arguments are of prime importance in exploration and the design space should accommodate and expose them. The primacy of self-generated representations in the form of free-hand sketches and the role of arbitrary visual stimuli as supporting design reasoning is introduced. The expositions lead to the conclusion that the design space should be conceived as a multilevel and multifacet construct that supports on the spot experimentation and provides essential feedback not only regarding designs, but also concerning the process of designing.