On the role of computational support for

designers in action
Marcelo Bernal, Department of Architecture, Universidad T
ecnica Federico
Santa Mar
ıa, Avenida Espa~na 1680, Valpara
ıso, 239-0123, Chile and College
of Architecture, Georgia Institute of Technology, Atlanta, GA 30332, USA
John R. Haymaker, College of Architecture, Georgia Institute of Technology,
Atlanta, GA 30332, USA
Charles Eastman, College of Architecture, Georgia Institute of Technology,
Atlanta, GA 30332, USA and College of Computing, Georgia Institute of
Technology, Atlanta, GA 30332, USA

Designers’ actions are high-level mechanisms based on heuristics and

assumptions learned from professional experience. Significant research has been
devoted to understanding these actions as well as finding ways to aid, automate,
or augment them with computational support. However, representing and
manipulating such tacit knowledge in computational environments remains an
open area of research. In this paper, we map designers’ actions and relationships
to compare them with computational approaches for the generation, evaluation,
and selection of design alternatives, and attempt to integrate all of the above.
The analysis provides a more thorough understanding of the role of
computational approaches in supporting designer actions and identifies
challenges and areas of future research.
Ó 2015 Elsevier Ltd. All rights reserved.

Keywords: design knowledge, computer aided design, design automation,

computer supported design, design technology

S
eminal studies on the behavior of designers in action (Sch€ on, 1983),
design thinking (Rowe, 1991), the mechanisms of knowing of the
design discipline (Cross, 2001), and the notion of design expertise
(Akin, 1988) have contributed to the understanding of the nature and
complexity of the mental processes of designers. Currently, we acknowledge
several well documented distinctive common actions that designers perform
in the act of designing, including interpreting design situations (Gero,
1998), co-evolving problems and solutions (Dorst & Cross, 2001; Maher &
Poon, 1996), recalling patterns of organization (Lawson, 2004), storing and
reusing expert knowledge from specific design domains (Moreno et al.,
Corresponding author: 2014; Popovic, 2004), and dividing tasks in distributed cognitive systems
Marcelo Bernal (Hollan, Hutchins, & Kirsh, 2000). All of them are iteratively executed during
marcelo.bernal@usm. the design process.
cl
www.elsevier.com/locate/destud
0142-694X Design Studies 41 (2015) 163e182
http://dx.doi.org/10.1016/j.destud.2015.08.001 163
Ó 2015 Elsevier Ltd. All rights reserved.
 Parallel to these efforts, research in computer-aided design (CAD) has evolved
from a focus on computer-aided drafting and modeling to the idea that com-
puters can aid these actions by manipulating abstract symbolic structures
similar to those created by the human mind (Kalay, 1985, p 321). Along with
observing increased computational power, we have witnessed the development
of several computational approaches addressing the entire cycle of the genera-
tion, evaluation, and selection of design alternatives. Such approaches range
from simply assisting to fully automating or even augmenting the actions of de-
signers and impact the efficiency and effectiveness of design exploration.

When characterizing the role of computers, we need to realize that what is in the
mind of the designer and what is represented in the computer are not the same.
Eastman (2001, p 6) states that the real structure supporting the design task is
an internal representation in the mind of the designer and that external repre-
sentations are auxiliary structures. The formalization of a design through a
model does not necessarily correspond to the complexity of the entire design it-
self. In fact, computer programs are integrated in more complex cognitive sys-
tems. The notion of distributed cognition developed by Hutchins (2000) while
he was studying navigation tasks describes these systems, which include the in-
teractions of internal and external representations with team members and even
with cultural contexts. Therefore, the challenge from the perspective of compu-
tational tools is supporting the behavior of designers within the larger systems
of interactions instead of simply reproducing their internal mental mechanisms.

One should also not disregard the threat of potential negative effects derived
from the use of computer programs. Robertson and Radcliffe (2009, p 137)
documented three such effects: Circumscribed thinking, or the limitation of
design alternatives to what can be done with the specific tool; premature fixa-
tion, or the resistance to making design changes resulting from the premature
complexity of the structure of the models; and bounded ideation, or the distrac-
tion from actual creative tasks resulting from technical and software issues
derived from the abuse of CAD tools. All three apparently detract from the
exploration of design alternatives.

While computers facilitate how designers manipulate information and perform

evaluations, limited evidence has also supported their role in improving design
quality in the creative process. Computer programs, although powerful tools,
also have several limitations that stem from their use of hierarchical data struc-
tures for representing geometry and related attributes such as boundary repre-
sentation (B-Rep) or constructive solid geometry (CSG) (Kalay, 1989), which
require explicit declarations that are frequently not available in conceptual
design stages, as Gross (1996, p 168) pointed out. In addition, computers are
limited at supporting changes in the topology of models that are beyond the
scope of dimensional variations, performing rapid evaluations based on heuris-
tics with partial information, and representing the diverse nature of the

164 Design Studies Vol 41 Part B November 2015

 elements of a design domain in the same environment. In addition, most design
tools have a significant cognitive cost (Eastman, 2001, p 22). Designers spend a
considerable amount of time attempting to interface their work rather than
focusing on the design itself, and as recent studies (Lawson & Dorst, 2009)
have pointed out, the fundamental characteristics of designers, such as their
tendency to rely on their experience and instincts to make decisions, cannot
currently be implemented within computational environments, so we should
consider examining less ambitious roles for a computer in design tasks.

Although reasonable skepticism exists, the aim of this study is to provide a

clearer understanding of the potential roles of computers in the design pro-
cess, to identify some of the circumstances in which computers are effectively
contributing to this process, and to visualize future areas of research that
could further support designers’ needs. To this end, this study compares
the well-documented characterization of designers in action mainly from
journals focusing on design activities with design computing-oriented jour-
nals that study computational approaches to the generation, evaluation,
and selection of design alternatives. This comparison attempts to characterize
the role of computational approaches related to actions performed by de-
signers in terms of identifying mainly human-based processes, computer-
aided processes that provide partial support to designers, automating pro-
cesses and turning them computer-based, or computer-augmented that
extend the designer compatibilities. The study identifies open areas of
research on computational support for designer in action and identifies
emerging approaches in systems engineering as a promising environment
for integrating various sources of knowledge.

1 Designers’ actions
Analyses of designer behavior during the design process have identified
various levels of expertise that depend on their background and experience:
novice, beginner, competent, expert, master, and visionary (Crismond, 2001;
Dreyfus, 1997; Lawson & Dorst, 2009). This study focuses on actions per-
formed on the expert level. Although expert designers have distinctive ap-
proaches to design, they share common characteristics. According to
Dabbeeru and Mukerjee (2008), they identify positive aspects of the design
process and immediately recognize when their interpretation of a problem is
accurate without extensive evaluation or further analysis. This apparent intu-
ition is derived from previous experience that developed critical thinking skills,
enabling them to assess the viability of potential design solutions within a very
short time. That is, designers intuitively understand, in advance, the many con-
straints that help them frame the problem and rapidly construct a design space
of feasible solutions. This section describes many of the common actions from
the literature such as framing design situations, formulating and evolving
design problems, recalling abstract patterns of organization, generating solu-
tions, and reusing knowledge. Figure 1 diagrams these actions and many of

Computational support for designers 165

Figure 1 Interaction of designers’ main actions, documented in an activity diagram in the System Modeling Language, SysML

their interdependencies, which will be discussed in the remaining sub-sections.

The diagram uses the System Modeling Language (SysML) (Friedenthal,
Moore, & Steiner, 2011), which is an object oriented language that represents
activities in terms of actions (rounded boxes) that take inputs and provides
outputs (square boxes), send signals triggering other actions (dashed lines),
and exchange information (continuous lines) while decisions making occurs
(rhomboidal shape).

1.1 Re-interpretation of design situations: reformulation,

analogies, and emergence
Studies on the interpretation of the design situation (Maher, Poon, &
Boulanger, 1996) show that re-interpretations alter the evolution of a design.
That is, a novel interpretation often produces a new representation that fosters
further interpretations, adding value to a design. Sch€
on (1983, p 49) referred to
this dynamic process as ‘reflection-in-action’, since new features, properties,
and other relevant aspects not intentionally planned for can emerge and affect
the evolution of the design. Although his study was based on sketches, it re-
mains valid for computer programs in which designers can recognize, detect,
or discover relevant elements of a situation or outcomes from their own oper-
ations that contribute to building re-interpretations, which rely on at least
three main actions: reformulating the structure, behavior, or function of the
design (Gero & Kannengiesser, 2004); building analogies that recall informa-
tion from previous experience or even other fields to address the current design
situation (Goel, 1997); and looking for emergent features that may be of further

166 Design Studies Vol 41 Part B November 2015

 use in the design process (Eastman, 2001; Gibson, 1977). These highly sophis-
ticated actions significantly influence the interpretation of the situation, and
questions regarding how they work and consequently how they can best be
supported by computers remain unanswered.

1.2 Formulation of design problems: framing, ill-definition

and Co-evolution
Expert designers usually formulate complex problems from initial ambiguous
or incomplete requirements. They speculate and conjecture about possible so-
lutions without expending too much effort analyzing the problem and instead
constantly shifting their goals and constraints (Cross, 2004). Since prefer-
ences or skills determine the aspects of the problem designers select to
work with, this behavior generates diverse problem formulations. Problem
formulation relies on actions that include framing the focus of interest
(Cross, 2004), building ill-defined problems (Dorst, 2007) and co-evolving the
problem with the solution (Maher & Poon, 1996). Framing the problem im-
plies setting the boundaries of the design situation, selecting the focus of
attention, and imposing coherence in decisions. This action not only occurs
during the initial stage, but also periodically along the entire design process.
Building ill-defined problems preserves the openness of the process. Some as-
pects of the problem are unavoidable and highly determined, other aspects
are under-determined and require interpretation, and other aspects are unde-
termined and leave room for the preferences of the designer who defines the
criteria to address them. While co-evolving problems and solutions, designers
use conjecture and tentative solutions as means to better understand the na-
ture of the problem. Tentative solutions often expose hidden aspects and
trigger the redefinition of the problem, which implies that the solution
must be adapted to new conditions. This dialog between problem and
solution iterates many times before reaching a matching problem-solution
pair. In fact, expert designers design not only the solution but also the
problem.

1.3 Recalling patterns of organization: chunks of constraints,

conceptual structures, and design schemas
The specific patterns that designers recall are abstract structures such as
chunks of constraints, which capture key tradeoffs of the problem
(Dabbeeru & Mukerjee, 2008), conceptual structures, which provide meaning
to the arrangements of parts from the perspective of the various systems
(Lawson & Dorst, 2009), and design schemas, which bring coherence and
identity to the overall design (Lawson, 2004). By recalling a chunk of con-
straints, designers know in advance many issues and conflicts that help
them frame the problem and then construct a design space with feasible so-
lutions to search for (Clevenger & Haymaker, 2011). Most of these con-
straints are often implicit relations that describe important aspects of the
problem such as the dependency among the geometry, fabrication

Computational support for designers 167

 limitations, cost, and timing. Therefore, techniques for splitting the problem
in chunks (Gobet et al., 2001) almost immediately build an interpretation and
provides clues about further steps.

Designers also describe what they know in terms of concepts rather than phys-
ical components. Such conceptualization is what defines the meaning of the el-
ements they use from the perspective of the various aspects of the design. For
example, architectural concepts related to aspects of accessibility such as hall-
ways, access, or corridors are abstract conceptual structures that are literally
not a single physical component or assembly. On the contrary, they are design
concepts that provide meaning to the arrangement of physical parts such as a
wall from the perspective of the accessibility system. Furthermore, the same
wall could also belong to the structural system. The complexity arises when
the designer constantly and unconsciously switches from the perspective of
one system to another during the design process. This oscillation helps to
define the physical components from multiple perspectives in a very synthetic
manner. By contrast, design schemas are neither conceptual structures such a
corridor nor a physical part such as a wall. They are an encrypted logic of or-
ganization that structures and gives identity to the design. According to
Lawson (2004, p 450), a design schema or schemata is a pattern of organiza-
tion rather than a typology. It includes both the geometry and all the related
relationships. A schemata probably represents one of the most relevant aspects
of a designer’s knowledge since it is a mechanism that organizes diverse infor-
mation about the design, embeds tacit knowledge representing the preferences
and guiding principles of a designer, and brings coherence to the design. How-
ever, how we might implement computational tools to support the mecha-
nisms behind the abstraction and recall of these patterns remains an open
question.

1.4 Generation of design solutions: parallel lines of thought,

knowledge integration, and preliminary evaluations
Studies on the notions of design problem and solution reveal that they are not
independent (Maher & Poon, 1996). Experts have the ability to conceptualize
the design situations, identify the underlying principles behind the problem,
redefine the problem, and reuse their experience to rapidly generate possible
matching solutions. Three actions that expert designer execute during the gen-
eration of such solution have been identified. They seem to be able to follow
parallel lines of thought by producing a range rather than a single solution, inte-
grate knowledge from different fields, and evaluate preliminary solutions (Cross,
2004). The ability to follow parallel options allows experts to keep their op-
tions open while the design proceeds. They explore the problem and solutions
in parallel, using a generative rather than a deductive approach (Rowe, 1991).
Testing options facilitates the exploration and identification of a match be-
tween problem and solution pair. To produce such set of solutions, experts
need integrate knowledge across different domains in parallel and

168 Design Studies Vol 41 Part B November 2015

 simultaneously perform preliminary evaluations based on mental simulations
(Dogan & Nersessian, 2010). These mechanisms enable expert designers to
produce reliable solutions early on. A major challenge for computational ap-
proaches, therefore, is supporting a divergent early design process instead of a
convergent one.

1.5 Reutilization of explicit knowledge from design domains:

recognizable problems, repository of parts, design rules and
evaluation methods
Expert designers rely on the reutilization of knowledge accumulated through
professional experience to respond rapidly and effectively. A design domain
groups and organizes knowledge about entities and relationships that describe
the universe of elements of the area of concentration that are frequently recalled
along the design process. The first step for further specialization and reutiliza-
tion of knowledge in a design domain is the identification and description of
primitives, or fundamental units in a domain (Sch€ on, 1988, p 189). These prim-
itives are typical design problems that suggest possible solutions, reusable and
adaptable physical parts, and rules distilled from heuristics or norms as well as
the explicit declaration of evaluation methods for assessing all kinds of aspects
of designs. Currently, many of these repositories of domain knowledge are sup-
ported by different but not necessarily integrated computational tools.

2 Computational approaches
Computational approaches support many designer actions. This section de-
scribes prevalent approaches, categorizes them into four focus areas: solution
generation, solution evaluation, decision-making, and attempts at integrating
all of them. The purposes, logic, advantages, and limitations of each approach
with respect to designer actions are presented.

2.1 Generation
Attempts to address the generation of design alternatives range between assist-
ing and trying to automate designers’ actions. Most approaches have been
originally conceived in engineering contexts, and adopted in industrial design
and architecture.

Parametric modeling computationally represents geometric relationships

organized in a hierarchical binary tree structure that is automatically updated
and visualized on the screen when changes in the parameters values occur
(Kalay, 1989). It captures dependencies among parameters, constraints, condi-
tionals, attributes and functions in order to control features of single parts and
assemblies of parts to drive the generation of tridimensional geometric models
(Nassar, Thabet, & Beliveau, 2003). Parametric modeling can embed success-
ful original designs representing the best design practices (Eastman, Teicholz,
Sacks, & Liston, 2011, p 241) and facilitate the design exploration based on
dimensional variation within the topological structure of a design

Computational support for designers 169

 configuration. Although it facilitates design reusability, parametric modeling
has limitations in terms of variation beyond its preconceived scope, prema-
turely reducing the range of options to explore.

Expert systems apply existing knowledge of design domains in similar situa-

tions emulating human decision making based on rules. Although they take
advantage from parametric technologies, they are beyond parametric libraries
since they do not only define the objects, but also the decision rules, the prob-
lem solving functions and tolerances to generate custom instances of the ob-
jects of the library according to various conditions (Eastman, Sacks & Lee,
2003). They capture in parametric libraries specific domain knowledge for de-
tailing and specification and usually use tridimensional conceptual models as
input geometry to trigger the propagation of parts (e.g. Structure, MEP, or
HVAC). However, expert systems, by definition, are limited to already known
solutions.

Case-based reasoning in design reuses and adapts previous solutions for

similar new problems. The problem is defined as a set of constraints that are
satisfied in order of importance. The systems record the solution and use it
as starting point for similar problems in the future. According to Kolodner
(1992, p 5) the effectiveness of this approach depends on the accumulated expe-
rience or cases, the ability to connect new situations with previous ones based
on similarities, the ability to adapt or combine old solutions to new problems
and the ability to adapt the evaluation of the outcomes in order to avoid mak-
ing the same mistakes by indexing the situation based on failures in the past.
These mechanisms also save time finding a solution by allowing the rapid gen-
eration of plausible ones, even though they need further validation. Case-
based reasoning facilitates reusability and continuous growth of the library
of solutions. However, while they apparently remember, they do not learn
and have issues synthesizing a large number of constraints for complex prob-
lems (Maher & Pu, 2014).

Generative design is an iterative cycle of rule application and evaluation.

Different from parametric modeling, the resulting design evolves from an
initial state through iteration by operations that imply geometric or topolog-
ical transformations. Examples can be found in genetic algorithms (Frazer,
Frazer, Liu, Tang, & Janssen, 2002) or shape grammars based on graphs
data structures (Grasl & Economou, 2013). Generative systems produce large
numbers of iterations in a short period of time generating several variations
based on the iterative re-adjustment of the parameter values and rules. Gener-
ative design can lead to apparently creative outcomes since every new combi-
nation of parameters brings the opportunity to look for the emergence of new
properties or affordances from the resulting composition. Although it is an
approximation to the co-evolving dialog between problem and solution that
characterizes expert designers, it lacks reformulation mechanisms.

170 Design Studies Vol 41 Part B November 2015

 Design languages enable the production of alternative design configurations
by combining well-known physical parts using design rules. The physical parts
are the primitive elements of the language, and are used to represent static as-
pects of the design. The rules represent the dynamic aspects since they provide
the main driving parameters and embed the procedural knowledge regarding
how to assemble the parts. Design languages are valid in restricted design do-
mains that have a limited set of elements and well-defined relationships among
them. Commercial examples can be found in the automobile and aerospace in-
dustry (B€ ohnke, Reichwein, & Rudolph, 2009; La Rocca, 2011; McCormack,
Cagan, & Vogel, 2004) and academic examples in shape grammars (Grasl &
Economou, 2013; Stiny, 1994). However, the a priori definition of parts and
rules is labor intensive and restricted to limited well-known design domains.

Agent-based design relies on the interaction of active and autonomous entities

with discrete information through iterations, and it is based on the notions of
agent and a neighborhood populated by other agents. Every agent has appro-
priate rationale for every step it executes (Modi, Tiwari, Lin, & Zhang, 2011).
The behavior of the agent can be driven by proactivity and cooperation among
agents with a common goal, or reactivity to changes in the environment. Such
autonomous reactions trigger the emergence of an apparently intelligence
without central reasoning control (Bento & Feij
o, 1997). Studies have explored
form finding methods embedding material, fabrication and geometric con-
straints (Baharlou & Menges, 2013), and also multi-objective optimization
problems in ship design by capturing various tradeoffs across chunks of tech-
nical constraints (Cui & Turan, 2010). The resulting outcome of all these
discrete interactions through time can eventually produce unexpected remark-
able results or fall into constraint circularity and non-sense loops.

2.2 Evaluation
Solution evaluation approaches analyze solution candidates and provide feed-
back to support decision-making. The notion of a computer as a design critic
refers to a wide variety of types of estimations, analyses and performance
evaluation.

Performance evaluations estimate how a product executes a given function un-

der stress. The performance-based design framework (Becker, 2008) focuses
on what the product should achieve instead of how it should be built. These
approaches depend on project goals and requirements which must be trans-
lated into quantitative indicators. In order to evaluate the degree of satisfac-
tion of such requirements different methods have been developed
(Augenbroe & Park, 2005). Performance evaluation for concerns including en-
ergy efficiency, structural performance, natural lighting, heat transfer, and
ventilation capture and reuse mostly explicit knowledge and provide feedback
for decision making. These approaches are difficult to execute in early design
stages because of ambiguity and incompleteness in the design definition.

Computational support for designers 171

 Rule checking evaluates the fulfillment of rules and provides feedback
regarding conflicts in four phases: rule interpretation, model preparation,
rule execution and results reporting. The rules represent critical requirements,
guidelines or norms. Rule-checking systems report conflict areas and provide a
reference to the violated rules. Such feedback is very valuable in large projects
where it is difficult to track all the rules by human reviewers (Eastman, Lee,
Jeong, & Lee, 2009, p 1014). However, the designer must interpret the reports
and provide solutions. Determining how to make or suggest corrections in real
time during the design process remains a challenge.

Constraint-based approaches capture design knowledge in the form of con-

straints and requirements that must be satisfied by the design. They represent
standard or custom conventions and norms that users assign to objects to be
combined. Examples of this approach can be found and in the implementation
of algorithms to search through design spaces (Thornton, 1996) and in tools
for floor plan layouts based on dimensioning and adjacencies that builds a sce-
nario to explore trade-offs among constraints (Vijayan & Tsay, 1991). If con-
straints are reached or broken immediate feedback is provided, supporting the
coevolution of the problem and the solution. However, constraint-based
methods only detect when constraints are violated and do not provide clues
about how to address the problem.

2.3 Selection
Decision making methods attempt to reflect trade-offs, values and preferences
through quantitative indicators complimented with relative weights to reflect
issues of certainty of information and preferences of decision makers.

Utility methods attempt to formally describe the preferences of the decision

maker. Choosing by advantages (CBA) makes the relevant attributes and ad-
vantages of the solution candidates explicit (Suhr, 1999). Collaborative weight,
rate and calculate (WRC) methods model the varied preferences amongst
stakeholders (Haymaker, Chachere, & Senescu, 2011). However, implementa-
tions of CBA and WRC are limited in capturing stakeholder utilities because of
their inability to consider uncertainties and their need for explicit information
that is not always available in the early design stage to support preliminary
evaluations. Addressing multiple criteria and stakeholders and accurately
capturing decision makers preferences remain active areas of research.

Multi-criteria decision approaches establish metrics to evaluate how a design

fulfills given objectives. Qualitative requirements are translated into measur-
able quantitative values, and acceptable limits of fulfillment. These approaches
aggregate data into results to measure performance, which may also be
normalized to establish a level of comparison of different design attributes
(Augenbroe & Park, 2005). Multidisciplinary performance analysis ap-
proaches provide flexibility to measure fulfillment of multiple requirements

172 Design Studies Vol 41 Part B November 2015

 by design alternatives, but because the comparison is based on quantitative
values, it is difficult to apply such methods in early design stages when the pre-
liminary evaluations occur.

Optimization methods developed in Mathematics and Operations research

enable designers to define and search through large spaces of designs.
Modeling and optimization efforts have increased exponentially with linearly
increasing input and output parameters. Therefore, in addition to parallel
computing and increasing computer power, researchers have generated a
wide array of methods including decomposing design problems into sub-
problems, screening significant variables, reducing design space, mapping,
and visualization (Kleijnen, 1997; Shan & Wang, 2010). While the iteration
process facilitates knowledge integration and negotiation between design
and engineering expertise during the design process and allows the visualiza-
tion of the trade spaces (Flager, Welle, Bansal, Soremekun, & Haymaker,
2009), standard optimization approaches require a high level design specifica-
tions before design decisions can be made.

2.4 Integration
Although generation, evaluation and selection correspond to different aspects
of the design process, the current focus is to integrate them into common en-
vironments to facilitate their interaction. Three main approaches exist for inte-
gration: custom system-to-system, interoperability based on information
exchange and the integration of different systems in the same environment.

Custom integration links generation, evaluation and selections tools or

methods in the same environment, usually for early design stages. Research ef-
forts linking analysis packages with conceptual tridimensional models of
buildings to eliminate file exchanges and improve the interactivity between
design, evaluation and selection (Sanguinetti, Bernal, El-Khaldi, & Erwin,
2010). Such custom integration is based on the implementation of one-to-
one selective data exchanges to provide an integrated environment to visualize
trade-offs among variety of requirements. The linkages are labor intensive in
terms of implementation and with limited reusability because of the specificity
of the integration.

Interoperability involves exchanges among systems that are based on industry

standard neutral file formats. Interoperability allows knowledge integration
regarding design, performance analyses, cost estimations, fabrication, sched-
uling, clash detection, rule-checking, and visualization. Examples can be found
in the Building Information Modeling (BIM) paradigm in the Architecture,
Engineering and Construction (AEC) Industry (Eastman et al., 2011) and
also in the STEP standard for engineering design (Eastman, 1999; Pratt,
2001). Although the exchanges are focused on the descriptions of physical
parts and their attributes, recent research efforts filter and exchange groups

Computational support for designers 173

 of components and attributes that belong to the same system of interest or
view of the model such as structure, energy consumption or costs. Similar to
the notion of conceptual structures those semantic exchange modules, or
SEMs, represent aspects that are distributed across physical parts
(Venugopal, Eastman, Sacks, & Teizer, 2012). Beside the progress in identi-
fying these abstract structures, current exchanges poorly preserve parametric
features and have not yet attempted to represent the simultaneous interaction
of different sources of design knowledge while designing. In addition, novel
design approaches are not represented in the exchange standards.

Model-based System Engineering (MBSE) is a model-based approach for inte-

gration of systems. MBSE implements computational modeling techniques to
support the formalization of requirements, design, evaluation, verification and
validation of a product (Reichwein & Paredis, 2011). This approach is based
on object-oriented non-system specific languages such as Unified Modeling
Language (UML) or System Modeling Language (SysML) for representation
of design knowledge in terms of objects with attributes and activities. These
languages also facilitate the integration of computational systems with specific
purposes into larger and complex multipurpose systems (Steinberg, Budinsky,
Merks, & Paternostro, 2008). They can connect computational systems for
generation, evaluation and also custom implementations to automate
decision-making. MBSE also allows continuous growth of the systems
(Tapscott & Williams, 2008). However, the development and integration of
these models is labor intensive and often requires programming translators
or wrappers to interface with different systems dramatically reducing the uni-
verse of potential users.

The next section uses these reviews of designer actions and computational ap-
proaches to provide a clearer description of the actual contribution of com-
puters with respect to different actions, and to identify open and promising
research areas.

3 The role of computational approaches that support

design actions
Figure 2 characterizes the role of computational approaches described above
in their relationship with actions performed by designers. The literature review
revealed several main actions: human baseddlacking of assistance; computer-
aideddproviding valuable feedback or facilitating tasks; computer-base-
ddautomating processes; and computer-augmenteddextending the de-
signer’s compatibilities to potentially improve the design quality. Neither
this classification nor our discussion implies an order of preference or attempts
an exhaustive analysis of all possible combinations, but instead, the matrix cre-
ates an overview of relationships that will help us visualize tendencies and
formulate questions.

174 Design Studies Vol 41 Part B November 2015

 Expert’s actions Generation Evaluation Selection Integration

Performance Evaluation

Custom tool integration

Casa-based Reasoning
Parametric Modeling

Generative Design

Design Languages
Computational

Constraint-based
Expert Systems

Interoperability
Rule Checking

Multi-attribute
Approaches

Optimization
Agent-based

MBSE
Utility
Reformulating SFB
tacit knowledge

Situation Forming Analogies

Looking for Emergence
Framing Design Situation
Problem Co-Evolving Problem-Solution
Building Under-Determined Problems
Recalling Chunk of Constraints o

Patterns Recalling Conceptual Structures

Recalling Design Schemata
Following Parallel Lines of Thought
Solution Evaluating Preliminary Solutions
explicit knowledge

Integrating Knowledge
Recalling Recognizable Problems
Building Repository of Parts
Domain Applying Design Rules
Applying Evaluation Methods

Figure 2 Classification of designer’s actions based on the level of support provided by computational approaches: white: human-based, light grey:
computer-aided, dark grey: computer-based, and black: computer-augmented

3.1 Human-based actions

Recent studies (Goel, Vattam, Wiltgen, & Helms, 2012) have found that
although most designer actions have some kind of computational support,
those related to defining design situations and recalling patterns of organiza-
tion are mainly human-based and provide minimal support for the cognitive
needs of designers. Despite some research effort (Davies, Goel, &
Nersessian, 2009), reformulation and forming analogies related to the interpre-
tation of design situations lack support mainly because such actions are not
fully understood and computer programs do not demonstrate a human-level
ability to link apparently non-related entities and build logical interpretation
(Goldscmidt, 1988). Even though design languages and case-based reasoning
capture some of the geometric and other relationships useful for recalling
and adapting previous solutions, a human designer must carry out the actual
assessment and selection of these patterns.

3.2 Computer-aided actions

In addition to their capabilities of drafting and tridimensional modeling,
which are low-level tasks, computers have demonstrated an ability to partially
support the formulation of a design situation and a co-evolving problem-

Computational support for designers 175

 solution by providing all kinds of feedback and some level of manipulation of
the geometry in real time, but such actions are typically driven by designers’
abilities to synthesize and think critically (Cross, 1990). Generation-oriented
approaches incorporate parameters and constraints according to a gradual
process of framing the design situation, accept some degrees of building
under-determined problems in terms of under or unconstrained parametric re-
lationships, and facilitate the production of alternative geometric models
following parallel lines of thought. The main contribution of evaluation and
selection-oriented approaches is providing multiple performance assessments
and normalized comparisons of features of alternatives that allow evaluating
preliminary solutions. Such feedback is gradually appearing in the early design
stage because of the emergence of tools for evaluating conceptual designs.

3.3 Computer-based actions

Computers can follow algorithmic processes that evaluate and search design
spaces or perform heuristic searches based on external knowledge provided
by experts, thus narrowing down the search space to only feasible solutions.
Although taking advantage of these computational approaches to automate
creative tasks is a desirable target, computer-based automation in design
largely consolidates explicit domain knowledge into standard procedures.
Automation is concentrated in actions that capture and reuse procedural
knowledge related to producing solutions and reusing domain knowledge
rather than addressing more challenging actions such as building analogies
or co-evolving a problem-solution. Generation-oriented approaches mainly
automate well-defined domain knowledge in terms of applying design rules,
building repositories of reusable physical parts, and applying evaluation methods
during the actual generation process. Automated design is appropriate for
applying well-defined knowledge, specifically for detailing. Similarly,
evaluation-oriented approaches automate the application of evaluation
methods as standardized procedures that subsequently support the integration
of knowledge from myriad fields.

3.4 Computer-augmented actions

Despite some moderate success in enhancing computational support in the
design process specifically regarding augmenting the design space by looking
for the emergence of new design features derived from the use of iterative algo-
rithms, most augmentations in the literature relate to producing design solu-
tions and capturing and reusing knowledge from design domains. In terms
of solutions, the main achievements have been following parallel lines of
thoughts and integrating knowledge. Design languages and MBSE can expand
design space from the boundaries of parametric variations to a family of topo-
logically varying configurations based on a combination of parts that can
effectively support the exploration of parallel alternatives. In addition, optimi-
zation approaches, custom integration, and MBSE actively address the integra-
tion of criteria derived from multiple aspects of a design. Recent progress

176 Design Studies Vol 41 Part B November 2015

 focuses on capturing and reusing the explicit knowledge of design domains.
Parametric modeling, design languages, and MBSE facilitate building reposi-
tories of parts and support the reuse of already known solutions representing
best practices. Finally, the approaches that allow the application of design rules
and evaluation methods in real time dramatically enhance the reliability of the
choices and decisions of designers.

4 Conclusions
The results of this study suggest three implications. The most fundamental is
the need for a better representation of the tacit aspects of design knowledge
before any computational implementation; another is the need for a change
in focus from the modeling of a design to the needs of designers; and finally,
the systems integration approach may offer a potential platform on which
the interaction among the various forms of design knowledge can be
integrated.

4.1 Tacit qualitative knowledge beyond manipulation of

explicit knowledge
This study illustrates the current progress in research related to the automation
and augmentation of actions to manipulate explicit knowledge from design
domains to produce solutions. However, computation research has made little
headway in addressing the manipulation of tacit knowledge related to qualita-
tive aspects of a design task because of a lack of understanding and represen-
tations of these implicit assumptions embedded in design decisions, which has
created challenges for computational implementations. These limitations are
concentrated in actions that frame the focus of interest, trigger unpredictable
inferences, and provide coherence to a design. As human intelligence can inter-
pret and manage relationships that a computer cannot, these actions mainly
rely on mental rather than external representation. That is, designers are
able to rapidly identify relevant aspects of a problem and constantly shift
the direction of design development through reformulation, analogy, or co-
evolution but receive little or no computational support for such behavior
or a means of integrating qualitative knowledge that enhances design quality.

4.2 Migration from design-to designer-centric computational

tools
While physical parts are driven by parametric relationships and constraints,
conceptual structures arrange these parts and the design schema drive the
overall integrity of the design. These patterns of organization represent the un-
derlying relationships that designers establish across components and subsys-
tems of a design. Although the ability to recall and adapt such abstract entities
is an important designer skill, current computational tools are design-centric,
with interfaces from the perspective of the physical components, rather than
designer-centric, with a focus on supporting the actions that designers execute
while they manipulate the patterns that drive the arrangement of the parts.

Computational support for designers 177

 Addressing the manipulation of these patterns appears to be a key to facili-
tating the reusability of design knowledge. In this regard, we have noted
some progress in computational support for the manipulation of parametric
relationships and constraints, but conceptual structures and design schemas
remain challenges. However, before we can adequately address computational
implementation in these areas, we must be able to determine and explain the
nature of immaterial entities that encapsulate the most sophisticated knowl-
edge that expert designers use to provide coherence, integrity, and identity
to their designs. Having representations of these patterns is the first step to
manipulating them, building associations, facilitating topological transforma-
tions, and identifying their emergence during the design process.

4.3 Towards integration for augmentation

While design knowledge is internally represented and connected in the mind of
the designer, external representations appear in a number of file formats, and
they are not necessarily integrated or exchangeable. The interoperability
approach, which relies on data exchange, and MBSE methods, which rely
on system integration, address these problems in different ways. Although
interoperability supports later stages of design development by allowing a va-
riety of file exchanges that effectively inform the evolution of the design, inter-
operability approaches have limited impact on the conceptual design when
information is not well structured and most of the tacit knowledge is recalled.
MBSE, by contrast, has become a flexible means of communicating across dis-
ciplines using a central repository of diverse knowledge that can be declared in
various formats for multiple sub-systems. Figure 2 shows MBSE methods for
enhancing most explicit designer actions; thus, it is perhaps the most suitable
for supporting distributed cognitive systems addressing conceptual and crea-
tive design problems. The MBSE approach is based on three main principles
that can effectively contribute to the augmentation of designers’ actions:
abstraction, integration, and continuous growth of a reusable design knowl-
edge repository. MBSE advocates the modeling of domain knowledge
abstractly, independent from any computational tool, which contributes to a
clearer definition of knowledge in conceptual terms and avoids any references
to the complexity of specific computational representations. The principle of
integration that drives the MBSE approach enables the interconnection of
apparently disperse information and the inheritance of most achievements
by single systems. The last principle of continuous growth implies that
capturing design knowledge is not a unique task, but rather a continuous pro-
cess of refining and adding new solutions that gradually extend the scope of the
domain knowledge repository, much as the human designer acquires knowl-
edge and skills.

Acknowledgments
The authors wish to acknowledge the support provided by the Fulbright Ex-
change Visitor Program N G-1-12218 (http://www.fulbright.cl) and

178 Design Studies Vol 41 Part B November 2015

 MECESUP MECE 2 Program (http://www.mecesup.cl) from Chile and the
Digital Building Laboratory Research Project N 4955617 (http://
www.dbl.gatech.edu) at Georgia Tech, Atlanta, Georgia, USA.

References
€ (1988). Expertise of the architect. In M. D. Rychener (Ed.), Expert sys-
Akin, O.
tems for engineering design (pp. 171e196). New York: Academic Press.
Augenbroe, G., & Park, C.-S. (2005). Quantification methods of technical build-
ing performance. Building Research & Information, 33(2), 159e172.
Baharlou, E., & Menges, A. (2013). Generative agent-based design computation.
Paper presented at the eCAADe 2013: Computation and PerformanceePro-
ceedings of the 31st International Conference on Education and research in
Computer Aided Architectural Design in Europe, Delft, The Netherlands,
September 18-20, 2013.
Becker, R. (2008). Fundamentals of performance-based building design. Paper pre-
sented at the Building Simulation.
Bento, J., & Feij
o, B. (1997). An agent-based paradigm for building intelligent
CAD systems. Artificial Intelligence in Engineering, 11(3), 231e244.
B€
ohnke, D., Reichwein, A., & Rudolph, S. (2009). Design language for airplane
geometries using the unified modeling language.
Clevenger, C. M., & Haymaker, J. (2011). Metrics to assess design guidance.
Design Studies, 32(5), 431e456.
Crismond, D. (2001). Learning and using science ideas when doing investigate-
and-redesign tasks: a study of naive, novice, and expert designers doing con-
strained and scaffolded design work. Journal of Research in Science Teaching,
38(7), 791e820.
Cross, N. (1990). The nature and nurture of design ability. Design Studies, 11(3),
127e140.
Cross, N. (2001). Designerly ways of knowing: design discipline versus design sci-
ence. Design Issues, Vol. 17, 49e55.
Cross, N. (2004). Expertise in design: an overview. Design Studies, 25(5),
427e441.
Cui, H., & Turan, O. (2010). Application of a new multi-agent hybrid co-
evolution based particle swarm optimisation methodology in ship design. Com-
puter-Aided Design, 42(11), 1013e1027.
Dabbeeru, M. M., & Mukerjee, A. (2008). Discovering implicit constraints in
design. In Design Computing and Cognition’08 (pp. 201e220). Springer.
Davies, J., Goel, A. K., & Nersessian, N. J. (2009). A computational model of vi-
sual analogies in design. Cognitive Systems Research, 10(3), 204e215.
Dogan, F., & Nersessian, N. J. (2010). Generic abstraction in design creativity: the
case of Staatsgalerie by James Stirling. Design Studies, 31(3), 207e236.
Dorst, K. (2007). The problem of design problems. In N. Cross (Ed.), Expertise in
Design e Design Thinking Research Symposium 6. Sydney: Creativity and
Cognition Studios Press.
Dorst, K., & Cross, N. (2001). Creativity in the design process: co-evolution of
problemesolution. Design Studies, 22(5), 425e437.
Dreyfus, H. L. (1997). Intuitive, deliberative, and calculative models of expert per-
formance. Naturalistic Decision Making 17e28.
Eastman, C. (1999). Building product models: computer environments, supporting
design and construction. CRC Press.
Eastman, C. (2001). New directions in design cognition: Studies of representation
and recall. Elsevier.

Computational support for designers 179

 Eastman, C., Sacks, R., & Lee, G. (2003). Development of a knowledge-rich CAD
system for American precast concrete industry. Paper presented at the Associa-
tion for Computer Aided Design in Architecture (ACADIA), Muncie, Indiana.
Eastman, C., Lee, J.-M., Jeong, Y.-S., & Lee, J.-K. (2009). Automatic rule-based
checking of building designs. Automation in Construction, 18(8), 1011e1033.
Eastman, C., Teicholz, P., Sacks, R., & Liston, K. (2011). The evolution from file-
based exchange to building model repositories. BIM Handbook: A guide to
Building Information Modeling for Owners, Managers, Designers, Engineers,
and Contractors (2nd ed.). John Wiley & Sons Inc.
Flager, F., Welle, B., Bansal, P., Soremekun, G., & Haymaker, J. (2009). Multi-
disciplinary process integration and design optimization of a classroom build-
ing. Journal of information technology in construction, 14, 595e612.
Frazer, J., Frazer, J., Liu, X., Tang, M., & Janssen, P. (2002). Generative and
evolutionary techniques for building envelope design.
Friedenthal, S., Moore, A., & Steiner, R. (2011). A practical guide to SysML: The
systems modeling language. Elsevier.
Gero, J. S. (1998). Conceptual designing as a sequence of situated acts. In Artifi-
cial intelligence in structural engineering (pp. 165e177). Springer.
Gero, J. S., & Kannengiesser, U. (2004). The situated functionebehaviourestructure
framework. Design Studies, 25(4), 373e391.
Gibson, J. J. (1977). The theory of affordances. Hilldale, USA.
Gobet, F., Lane, P. C., Croker, S., Cheng, P. C., Jones, G., Oliver, I., et al. (2001).
Chunking mechanisms in human learning. Trends in Cognitive Sciences, 5(6),
236e243.
Goel, A. K. (1997). Design, analogy, and creativity. IEEE Expert, 12(3), 62e70.
Goel, A. K., Vattam, S., Wiltgen, B., & Helms, M. (2012). Cognitive, collabora-
tive, conceptual and creativedfour characteristics of the next generation of
knowledge-based CAD systems: a study in biologically inspired design. Com-
puter-Aided Design, 44(10), 879e900.
Goldscmidt, G. (1988). Interpretation: its role in architectural designing. Design
Studies, 9(4), 235e245.
Grasl, T., & Economou, A. (2013). From topologies to shapes: parametric shape
grammars implemented by graphs. Environment and Planning B: Planning and
Design, 40(5), 905e922.
Gross, M. D. (1996). The electronic cocktail napkinda computational environ-
ment for working with design diagrams. Design Studies, 17(1), 53e69.
Haymaker, J. R., Chachere, J. M., & Senescu, R. R. (2011). Measuring and
improving rationale clarity in a university office building design process. Jour-
nal of Architectural Engineering, 17(3), 97e111.
Hollan, J., Hutchins, E., & Kirsh, D. (2000). Distributed cognition: toward a new
foundation for human-computer interaction research. ACM Transactions on
Computer-Human Interaction (TOCHI), 7(2), 174e196.
Hutchins, E. (2000). Distributed cognition. International Encyclopedia of the Social
and Behavioral Sciences. Elsevier Science.
Kalay, Y. E. (1985). Redefining the role of computers in architecture: from draft-
ing/modelling tools to knowledge-based design assistants. Computer-Aided
Design, 17(7), 319e328.
Kalay, Y. E. (1989). Modeling objects and environments. Wiley.
Kleijnen, J. P. (1997). Sensitivity analysis and related analyses: a review of some
statistical techniques. Journal of Statistical Computation and Simulation,
57(1e4), 111e142.
Kolodner, J. L. (1992). An introduction to case-based reasoning. Artificial Intel-
ligence Review, 6(1), 3e34.

180 Design Studies Vol 41 Part B November 2015

 La Rocca, G. (2011). Knowledge based engineering techniques to support aircraft
design and optimization. TU Delft, Delft University of Technology.
Lawson, B. (2004). Schemata, gambits and precedent: some factors in design
expertise. Design Studies, 25(5), 443e457.
Lawson, B., & Dorst, K. (2009). Design expertise. Recherche, 67, 02.
Maher, M. L., & Poon, J. (1996). Modeling design exploration as co-evolution.
Computer-Aided Civil and Infrastructure Engineering, 11(3), 195e209.
Maher, M. L., Poon, J., & Boulanger, S. (1996). Formalising design exploration as
co-evolution: A combined gene approach.
Maher, M. L., & Pu, P. (2014). Issues and applications of case-based reasoning to
design. Psychology Press.
McCormack, J. P., Cagan, J., & Vogel, C. M. (2004). Speaking the Buick lan-
guage: capturing, understanding, and exploring brand identity with shape
grammars. Design Studies, 25(1), 1e29.
Modi, S., Tiwari, M., Lin, Y., & Zhang, W. (2011). On the architecture of a
human-centered CAD agent system. Computer-Aided Design, 43(2), 170e179.
Moreno, D. P., Hernandez, A. A., Yang, M. C., Otto, K. N., H€ oltt€
a-Otto, K.,
Linsey, J. S., Wood, K., & Linden, A. (2014). Fundamental studies in
design-by-analogy: a focus on domain-knowledge experts and applications to
transactional design problems. Design Studies, 35(3), 232e272.
Nassar, K., Thabet, W., & Beliveau, Y. (2003). Building assembly detailing using
constraint-based modeling. Automation in Construction, 12(4), 365e379.
Popovic, V. (2004). Expertise development in product designdstrategic and
domain-specific knowledge connections. Design Studies, 25(5), 527e545.
Pratt, M. J. (2001). Introduction to ISO 10303dthe STEP standard for product
data exchange. Journal of Computing and Information Science in Engineering,
1(1), 102e103.
Reichwein, A., & Paredis, C. J. (2011). Overview of Architecture Frameworks and
Modeling Languages for Model-Based Systems Engineering. Paper presented at
the Proceedings of the ASME 2011 International Design Engineering Tech-
nical Conferences & Computers and Information in Engineering Conference.
Robertson, B., & Radcliffe, D. (2009). Impact of CAD tools on creative problem
solving in engineering design. Computer-Aided Design, 41(3), 136e146.
Rowe, P. G. (1991). Design thinking. The MIT Press.
Sanguinetti, P., Bernal, M., El-Khaldi, M., & Erwin, M. (2010). Real-time design
feedback: Coupling performance-knowledge with design iteration for decision-
making. Paper presented at the Spring Simulation Multiconference, Orlando,
Florida.
Sch€on, D. A. (1983). The reflective practitioner: How professionals think in action,
Vol. 5126. Basic Books.
Sch€on, D. A. (1988). Designing: rules, types and words. Design Studies, 9(3),
181e190.
Shan, S., & Wang, G. G. (2010). Survey of modeling and optimization strategies
to solve high-dimensional design problems with computationally-expensive
black-box functions. Structural and Multidisciplinary Optimization, 41(2),
219e241.
Steinberg, D., Budinsky, F., Merks, E., & Paternostro, M. (2008). EMF: Eclipse
modeling framework. Addison-Wesley Professional.
Stiny, G. (1994). Shape rules: closure, continuity, and emergence. Environment and
Planning B, 21(7), 49e78.
Suhr, J. (1999). The choosing by advantages decisionmaking system. Greenwood
Publishing Group.

Computational support for designers 181

 Tapscott, D., & Williams, A. D. (2008). Wikinomics: How mass collaboration
changes everything. Penguin.
Thornton, A. C. (1996). The use of constraint-based design knowledge to improve
the search for feasible designs. Engineering Applications of Artificial Intelli-
gence, 9(4), 393e402.
Venugopal, M., Eastman, C., Sacks, R., & Teizer, J. (2012). Semantics of model
views for information exchanges using the industry foundation class schema.
Advanced Engineering Informatics, 26(2), 411e428.
Vijayan, G., & Tsay, R.-S. (1991). A new method for floor planning using topo-
logical constraint reduction. IEEE Transactions on Computer-Aided Design of
Integrated Circuits and Systems, 10(12), 1494e1501.

182 Design Studies Vol 41 Part B November 2015