Essays on a Novel Framework for Product Development Theory by Sourobh Ghosh B.S. Mechanical Engineering B.A. Economics University at Buffalo, The State University of New York, 2013 Submitted to the Department of Mechanical Engineering in Partial Fulfillment of the Requirements for the Degree of Master of Science in Mechanical Engineering at the ARCHIVES MASSACHUSETTS INSTITUTE OF TECHNOLOGY OCT 012015 Massachusetts Institute of Technology LIBRARIES September 2015 C 2015 Massachusetts Institute of Technology. All rights reserved. Signature redacted Signature of A uthor..................................... ....................................................... Sourobh Ghosh Department of Mechanical Engineering July 31, 2015 Signature redacted .................................................... Warren P. Seering Weber-Shaughness Professor of echanical Engineering and Engineering Systems Thesis Supervisor Certified by....... Signature redacted Accepted by....................................... ... . .... David E. Hardt Ralph E. and Eloise F. Cross Professor of Mechanical Engineering Graduate Officer, Department of Mechanical Engineering 2 Essays on a Novel Framework for Product Development Theory by Sourobh Ghosh Submitted to the Department of Mechanical Engineering on July 31, 2015 in Partial Fulfillment of the Requirements for the Degree of Master of Science in Mechanical Engineering Abstract After calls for a common terminology to unify the highly distributed literature on engineering design, this thesis develops and presents an emergent framework for the novel description and reconceptualization of the standard product development (PD) process. We begin with a review of the literature on Set-Based Design (SBD), introducing a paradigm called SetBased Thinking which unifies the SBD and affiliated literatures around common themes and influences. Motivated by a set-based perspective, we then establish the lexicon for a novel framework for PD. Using this lexicon to drive descriptions of PD, we gain new insights on the process, such as the relationship between function and form in early stage design and designers' reliance on articulating form in order to explore the functional space. These insights and others lead us to establish the Designer's Dilemma, which asserts that fixation is an inevitable consequence of exploring ideas in early stage design that cannot be avoided with current ideation techniques. In sum, this thesis articulates a framework which presents a fundamental reconceptualization of the front end of the PD process. We also identify future areas of work upon which the framework can be expanded to address latent research issues in the PD literature. Thesis Supervisor: Warren P. Seering Title: Weber-Shaughness Professor of Mechanical Engineering and Engineering Systems 3 4 Acknowledgements I shall begin by acknowledging sources of financial support for this thesis. This material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. 1122374. This work was also made possible in part by the Center for Clean Water and Clean Energy at MIT and the King Fahd University of Petroleum and Minerals (KFUPM) under project number Ri 1-DMN-09. There are countless individuals who have helped make this thesis a reality. While I unfortunately will not be able to recognize them all, I shall mention a few here. First, I would like to thank my advisor, Prof. Warren Seering, for his invaluable guidance and for embracing the unconventional path I took during the course of this thesis. Our conversations on product development, education, and an array of other topics have always been a source of inspiration. I am truly fortunate to have had a mentor of such high character oversee the most intellectually stimulating years of my life thus far. I owe special thanks to Maria Yang, David Wallace, and Steven Eppinger for supporting my endeavors and for their wonderful advice and feedback. I would also like to acknowledge my colleagues and friends on the fourth floor of Building 3. Alison, Elisabeth, David, Endris, and the folks at Ideation and CADLAB have made this experience so much more enjoyable and worthwhile. Thank you to Kemper Lewis and the DOES lab for introducing me to and sharing your passion for the world of product development research. Finally, this work would not have been possible without the support of friends and family. To my parents and brother, thank you instilling your love for learning upon me and for inspiring me to always challenge myself. These values and your continued support are the reason why I am here today. 5 6 Table of Contents 1. Introduction............................................................................................................................... 10 2. Set-Based Thinking in Product Developm ent......................................................................... 14 2.1 Introduction and M otivation .......................................................................................... 14 2.2 Review M ethodology...................................................................................................... 16 2.3 Seven Characteristics of Set-Based Product Developm ent.............................................. 18 2.4 The Two Principles of Set-Based Thinking .................................................................... 30 2.5 Case Exam ple: Set-Based Thinking in Function Structures ........................................... 37 2.6 D iscussion and Conclusion ............................................................................................. 38 3. A New Fram ew ork for the Product Developm ent Process.................................................... 41 3.1 M otivation and Background........................................................................................... 41 3.2 Definitions.......................................................................................................................... 43 3.3 Role of Term s in Fram ework Thinking .......................................................................... 54 3.4 U se in the Literature........................................................................................................... 54 3.5 Conclusion ......................................................................................................................... 60 4. D escribing the Process Using the Fram ework ...................................................................... 61 4.1 Tem poral D escription of the Product D evelopm ent Process .......................................... 61 4.2 Other Visual Representations of the Fram ework ........................................................... 63 4.3 Observations from Using the Fram ew ork ...................................................................... 65 4.4 Conclusion ......................................................................................................................... 66 5. Surjective Relationship ............................................................................................................. 67 5.1 The Surjective Theorem for Product D evelopm ent ........................................................ 67 5.2 Understanding the Dom ains........................................................................................... 69 7 5.3 Additional Insights from the Surjective Relationship........................................................ 71 5.4 Conclusion ......................................................................................................................... 73 6. The Fram ework: Ideal vs. Practice ....................................................................................... 74 6.1 An Idealized M odel for Product Developm ent ............................................................... 74 6.2 Product Development in Practice, Described by the Framework ................................... 75 6.3: Conclusion ........................................................................................................................ 79 7. The Designer's D ilem m a........................................................................................................ 80 7.1 The Designer's D ilem m a ................................................................................................. 80 7.2 Contextualizing the Designer's Dilem ma in the Literature ............................................ 82 7.3 An Exam ple from Football ................................................................................................ 83 7.4 Observations from Reasoning with the Designer's Dilemma.................... 85 7.5 Strategies to Counteract the Designer's Dilemm a .......................................................... 89 7.6 Conclusion ......................................................................................................................... 91 8. Understanding Constraint Propagation................................................................................. 93 8.1 Identifying an Order for Constraint Propagation ............................................................ 93 8.2 An Econometrics Inspired Framework for Ordering Constraint Propagation ............... 94 8.3 Conclusion ......................................................................................................................... 96 9. Bringing it Together: How the Framework Influences the Way We Think about Design....... 98 .......... ............ ......... . . 9.1 W hat Does This Work Hope to Achieve ...................................... 9.2 What N ew Ways of Thinking Are Proposed?.................................. 9.3 What are .............................................. 98 ....... ........... ........... . . .......... .............. .............. 9.4 Concluding Thoughts ....................................................................................................... 99 105 107 References................................................................................................................................... 109 Appendix..................................................................................................................................... 117 8 A l The Pareto Frontier as a Subset of the Surjection's Codomain ....................................... 117 A2 Cardinality of the Two Domains...................................................................................... 117 9 1. Introduction "Everyone designs who devises courses of action aimed at changing existing situations into preferredones." - H erbert A. Simon, The Sciences of the Artificial In Herbert Simon's influential work The Sciences of the Artificial, Simon describes an interesting trend in higher education over much of the 20th century, noting "...it is ironic that in this century the natural sciences almost drove the sciences of the artificial from professional school curricula..." (Simon 1996). Having established the study of design as a quintessential artificial science, Simon was particularly disturbed by his observation, especially in light of his view of design as "the core of all professional training." Simon argues that in order for engineering schools to deliver on their mission to train professionals, the study of design should be researched and taught as a science. He states, "The professional schools can reassume their professional responsibilities just to the degree that they discover and teach a science of design, a body of intellectually tough, analytic, partly formalizable, partly empirical, teachable doctrine about the design process." Many in the engineering community took note of Simon's arguments, triggering a number of activities that led to the creation of the Design Theory and Methodology (DTM) conferencea community of scholars passionately engaged in developing the science of design. At the 25th anniversary of this conference in 2013, a series of essays commissioned by the DTM Conference and Program Chairs explored past winners of the "Best Paper Award" in an effort to describe the development of past, present, and future research themes explored by the DTM community (Braha et al. 2013). This led a revisiting of Set-Based Design (SBD), a product development' (PD) 1 I use "product development" to describe the act of creating a product. This involves a host of tasks, including traditional engineering analysis, project management, marketing, finance, legal, to name a few activities. While Simon's definition of design (see above) reasonably allows "design" and "PD" to be used interchangeably, given notions in the broader literature which distinguish "design" from some of the aforementioned holistic tasks in bringing a product to market (such as considering marketing and legal concerns), I often elect in this work to use the broad implications of "product development" to refer to the process of product realization. 10 methodology that was formally introduced to the DTM community via Ward and Seering's 1989 DTM paper (A. C. Ward and Seering 1989). SBD is a fascinating example of a methodology which contributes to Simon's desire for developing design science. As a process, SBD meets many of Simon's criteria for design science, being "intellectually tough, analytic, partly formalizable, partly empirical." However, in the years since Ward and Seering's 1989 DTM paper, many of the authors who studied and applied SBD seemed to have differing views on how exactly to define the method. Thus, it clearly falls short of Simon's criteria for being a "teachable doctrine," since in order to be teachable, there must be clear consensus on how to define the method. Interestingly, as we discuss between the conclusion of Chapter 2 and the introduction of Chapter 3, there seems to be a lack of consensus on a variety of fundamental constructs within our theories for design and PD. Thus, this transcends SBD and is indicative of an issue in the greater design literature. Without a standardized teachable doctrine, we cannot realize Simon's vision for developing design into a comprehensive science. To appreciate this issue, one may review a nonexhaustive sampling of texts on design and the PD process (Hubka, Andreasen, and Eder 1988; Pahl et al. 2007; Pugh 1991; Suh 1995; Ulrich and Eppinger 2012). While the basic chronology of events describing the process remains roughly the same across these texts, there are fundamental incompatibilities among many of these works. For instance, while Pahl and Beitz describe an intermediary "Embodiment" phase to the process, their contemporary Pugh cannot easily accommodate this phase in his framework for Total Design, which goes straight from Concept Development to Detail Design. This thesis is my attempt to propose what could hopefully inform a teachable doctrine for design science. I do this by introducing a novel framework for PD theory. I begin by introducing base definitions which form the lexicon for the framework. This lexicon is designed with the intent to be general enough to describe a variety of scenarios within the PD process while being detailed and precise enough to guide deeper insights. With this carefully standardized terminology, I use this emergent framework to describe the process and make deeper observations about the underlying mechanics of reasoning throughout the PD process, particularly in the ill-defined, enigmatic front end of the process. 11 Each of the following chapters can be thought of as short essays which contribute to the development of this novel framework. An analogy can be drawn between the development of this framework and common law. In a common law system, the legal framework is decidedly emergent, whereby the principle of stare decisis establishes precedent of past decisions ("Stare Decisis" 2015). This indirectly implies that future decisions, often referencing past decisions, give greater meaning to these past decisions. Consider the example of the Fourteenth Amendment to the US Constitution. While originally proposed to grant citizenship to freed slaves after the Civil War, the Fourteenth Amendment has been used as the legal basis by which to decide cases on segregation in public schools, 2 abortion rights, and same-sex marriage, to name a few examples ("Fourteenth Amendent of the United States Constitution" 2015). Similarly, the framework for PD that I propose can be thought of as an emergent framework, where each subsequent chapter in this thesis expands upon and gives greater meaning to the existing framework formally established in each chapter preceding it. This implies that by the end of this thesis, the framework will not be complete in any conventional sense, inviting ample opportunity for future work. The remainder of the thesis is organized as follows: " Chapter 2 reviews the SBD and affiliated literature in order to establish Principles for SetBased Thinking and root our development of the framework in the chapters to follow. - Chapter 3 formally introduces the terminology of the framework and compares my use of terminology to common established uses in the literature. " Chapter 4 uses the framework to guide descriptions of the PD process, introducing critical observations that form the basis of developments in Chapters 5-8. " Chapter 5 formalizes the relationship between architectural and specification domains as first presented in Chapter 3 via the Surjective Theorem. The implications of this Theorem on the framework presented thus far and on other streams of work are also discussed. 2The case of Brown v. BoardofEducation (1954) is interesting as it specifically overturns a previous Supreme Court ruling, Plessy v. Ferguson (1896), thus refuting legal precedent. I point it out here to complete my analogy of developing an emergent framework for PD, since should any content within the framework be found to be incorrect after the publication of this thesis, it should be pointed out and corrected in future work. 12 " Chapter 6 presents an idealized model for PD based on the framework, which is then used in contrast to illustrate the realities of PD in practice. These insights help contextualize the articulation of the Designer's Dilemma. m Chapter 7 introduces the Designer's Dilemma and explores its implications for early stage design and ideation in general. It also explores strategies to attempt to counteract the effects of the Dilemma. " Chapter 8 proposes the creation of a framework for ordering the propagation of constraints by leveraging techniques from econometric analysis. " Chapter 9 collects and summarizes major results from reasoning with the framework over the previous chapters. In doing so, it articulates potential academic contributions while identifying areas for future work. 13 2. Set-Based Thinking in Product Development In the previous chapter, I establish the context of revisiting influential PD methodologies in the years since the establishment of the DTM conference, including Set-Based Design (SBD). After a series of academic case studies had revealed Toyota's unique product development practices to the world, a flurry of research has been conducted into SBD. In this chapter, we review work related to set-based design across academic communities in efforts to find common themes and influences. After a review of this literature, we inductively arrive at two Principles of SetBased Thinking: considering sets of distinct alternatives concurrently and delaying convergent decision making. These Principles allow us to articulate a working description of SBD, while presenting vital context for us to develop the novel framework for PD that is the primary focus of the remainder of this thesis. 2.1 Introduction and Motivation Set-Based Design (SBD), also referred to as Set-Based Concurrent Engineering (SBCE), has played an important role in the development of the DTM and the greater engineering design community, as established in 2013's retrospective study of DTM best papers and underlying research themes (Braha et al. 2013). The theoretical foundation for SBD was laid in Allen Ward's PhD thesis (A. C. Ward 1989), the fundamental ideas of which he presented to the Design Theory and Methodology (DTM) community in (A. C. Ward and Seering 1989). Ward's work nominally describes a computer program that selects standard components from catalogs in order to implement a variety of mechanical designs (A. C. Ward and Seering 1989). However, the fundamental idea behind this work was that products should be designed with all viable options in mind, and that options should not be eliminated unless there is a logical reason to do so. This proposed approach of exploring concepts in parallel while gradually narrowing the solution space until a single solution is found was in contrast to a traditional, pointbased design (PBD) approach which advocates generating alternatives at the onset and selecting only one for further development (Sobek, Ward, and Liker 1999; Singer, Doerry, and Buckley 2009). Thus, to keep all feasible options in play for consideration for as long as reasonable-this 14 embodies the ethos of SBD. The way Ward achieved this was by reasoning with sets, leading to the name Set-Based Design. While these ideas provided a fresh take on PD practice when they were introduced, the idea of considering design concepts in parallel was not new. Pahl and Beitz (Pahl and Beitz 1988) and Hubka (Hubka, Andreasen, and Eder 1988) described design processes for which alternate design concepts were considered simultaneously. The influence of processes associated with the consideration of sets of design alternatives was extended in the mid-1990's, when Ward and colleagues studied new PD at Toyota (A. Ward et al. 1995). At the time, Toyota enjoyed a reputation for developing reliable, superior vehicles under short timelines and at low cost. During their study, Ward et al. found that Toyota's unique practices had embodied the principles of SBCE, or SBD, reasoning that these practices were fundamental to Toyota's success. The influence of Ward et al.'s study was widespread. This study, along with related work (Womack, Jones, and Roos 2007), contributed to an environment which fostered the lean manufacturing movement that continues to impact new PD today. Furthermore, the influence of the Toyota case studies generated a flurry of research in the engineering design and management science communities, from where the original ideas of SBD were formulated. However, beyond these native academic communities, SBD began to influence areas such as civil and structural engineering, ship-building, and software development, to name a few examples. Among these diverse applications, SBD as a research area began to take a life of its own, potentially meaning something slightly different to each of its varied practitioners. This chapter reviews literature that addresses reasoning about sets of designs in an effort to find common ideas and themes. As this work is still preliminary, the literature reviewed in this work is not exhaustive; rather, an emphasis was placed on demonstrating a breadth of work across domains. This includes reviewing 1) methods that are explicitly categorized by some to be SBD and 2) methods that have not been explicitly categorized as SBD but that enable decision-making or reasoning about sets of design alternatives, such as use of function structures and shape grammars. In this chapter we will refer to the employment of both categories of methods as "setbased thinking." 15 With the diversity of work that has emerged over the last approximately two decades, we have struggled in our search of the literature to find a consensus over what SBD or set-based thinking represents today. Therefore, in this work we hope to identify emergent principles which embody commonly espoused views regarding the core principles of set-based thinking from the literature across communities. This chapter is organized as follows. In Section 2.2, we discuss the research methodology for our study, outlining our inductive approach. In Section 2.3 we explore 7 Characteristics of SetBased Product Development, and in Section 2.4 we use these 7 Characteristics to arrive at 2 Principles of Set-Based Thinking: considering sets of distinct alternatives concurrently, and delaying convergent decision making. In Section 2.5 we perform a case example on function structures, observing the ways in which the 2 Principles manifest themselves in this popular theoretical construct, and in Section 2.6 we discuss the implications of this study and conclusions. 2.2 Review Methodology In this section, we discuss the theoretical framework for our study. As stated in the introduction, our goal in this work is to consolidate the diverse literature on SBD methods and techniques around common themes which illustrate set-based thinking. Due to the flurry of research over the last two decades that has led to the evolution of SBD as a topic across various domains, we have decided to take an inductive approach to our study. Our approach shall be to review work which nominally exhibits characteristics or specific instances of SBD or, more generally, set-based PD processes. Then, from a holistic analysis of the work reviewed from these Characteristics, we shall arrive at Principles of Set-Based Thinking. This is in contrast to a deductive approach, which would involve formulating the Principles first, and then testing these Principles across each documented case of set-based affiliated research for validity. Not only would such a study be practically infeasible, but it would suggest that the premises (i.e. the derived Principles) be necessarily true in order for any proposed argument to be sound (and thus to have any value). To avoid this intellectual trap, we use an inductive approach to arrive at our probable conclusions: the Principles of Set-Based Thinking. We believe this analysis to be appropriate for this initial study. 16 The first step in this process is to identify common characteristics of set-based product development. Given the role of Ward et al. (1995)'s study of Toyota's PD processes in motivating the push for SBD research (a fact which can be substantiated by the frequent citations of (A. Ward et al. 1995) by the research reviewed in this work), we believe that the case studies presented in the 1995 work are a good place to find and define these characteristics. After reviewing these case studies and selecting the most noteworthy, unique attributes of Toyota's PD practices as reported in (A. Ward et al. 1995), we arrive at 7 Characteristics of Set-Based Product Development (SBPD) 3 , summarized below: 1. Emphasis on frequent, lo-fidelity prototyping 2. Tolerance for under defined system specifications 3. More efficient communication among subsystems 4. Emphasis on documenting lessons learned and new knowledge 5. Support for decentralized leadership structure and distributed, non-collocated teams 6. Supplier/subsystem exploration of optimality 7. Support for flow-up knowledge creation The Characteristics, forming a notion of what research into set-based methods entails, act as a filter by which we choose what work to review in this study. Using this filter, we review two classes of literature since the Ward et al. (1995) study-work that is nominally classified as SBD, and work that is not nominally classified as SBD but that utilizes set-based reasoning. Both classes of literature are part of the narrative on the influence of set-based thinking in the design community and beyond. 3The difference between Set-Based Design (SBD) and Set-Based Product Development (SBPD) as we refer to them is nuanced. Where SBD refers specifically to methodology, SBPD refers to the practical situations and outcomes where SBD is applied. 17 2.3 Seven Characteristics of Set-Based Product Development This section is dedicated to exploring the manifestation of the 7 Characteristics of SetBased Product Development across the literature. We detail the genesis of the Characteristics from (A. Ward et al. 1995), in addition to how the literature since this work supports each individual characteristic. We begin with the first characteristic, an emphasis on frequent, lo-fidelity prototyping. 2.3.1 Emphasis on Frequent, Lo-Fidelity Prototyping At the time of Ward et al. (1995)'s case studies, Toyota was noted as a leader in lean manufacturing and was the industry standard-bearer for the most efficient PD processes, both in terms of time and cost. It was for this reason that one of the most counterintuitive revelations of Ward et al.'s studies was Toyota's practice of frequent prototyping. Ward et al. characterized Toyota's system as generating "excessive numbers of prototypes" when compared to other Japanese and US automakers. For instance, while another Japanese or US automaker may develop on average three 1/5 scale clay models of body designs, Toyota may develop anywhere from 5 to even 20 of such models. Frequent Prototyping These results generated interest in the value of prototyping in the design process, spurring further research. Yang recognizes prototypes as "early embodiment[s] of a design concept" (Yang 2005), although it should be noted that prototypes may be generated at later phases of the design, and may embody several forms, from sketches (Yang 2008; Shah et al. 2001), to 3D CAD models (Nahm and Ishikawa 2006), to full-scale models described in (A. Ward et al. 1995). Notably, the proliferation of prototyping throughout the design process is a clear manifestation of concurrent development, as multiple prototypes help designers explore multiple concepts-which Toyota clearly understood and practiced. As Yang notes, prototyping can be thought of as a "design language," or a "way to express design thinking" (Yang 2005)descriptions which support the understanding that prototypes enable the concurrent development 18 of concepts. This reasoning will be important later on when we formulate Principles of Set-Based Thinking. While Ward et al. (1995) document Toyota's correlation with frequent prototyping and ostensibly superior design outcomes (i.e. cheaply manufactured, high-quality automobiles), recent work from Haggman et al. seeks to explore this relationship in a series of controlled studies (Haggman, Honda, and Yang 2013). Their research demonstrates statistically significant correlations between prototyping early in the design process and design outcomes. Furthermore, they find no statistically significant correlations between the time that design teams spent developing their prototypes and outcomes. Therefore, their results advocate "prototyp[ing] cheaply and early in a project." While the discussion above has generally supported the value of frequent prototyping, the other half of our argument is that prototypes developed should not intentionally exhibit very high fidelity or levels of resolution or detail-encompassed by Haggman et al.'s suggestion to develop them "cheaply." Here, it is instructive to consider examples from software design. Prototyping with Sufficient Fidelity Today, the agile software development movement continues to influence and inform much of the software industry's practices. The relationship between the agile software development movement and lean manufacturing has been well documented, suggesting that lean principles were important to influencing agile PD philosophies (Poppendieck and Poppendieck 2003). A popular method emerging from this movement was and continues to be Scrum. In general, Scrum and other Agile methodologies emphasize the rapid development of software concepts for trial and error without excessive commitment from developers. These "prototype" phase user interfaces (Uls) need to be tested for basic functionality, bugs, and other issues prior to implementation of the final details of the UI (Sutherland et al. 2007). Similarly motivated practices are found at one of the world's leading design firms, IDEO. The prototyping philosophy at IDEO follows the three R's: rough, rapid, and right. This philosophy promotes the development of prototypes that have sufficient detail-not too much, but just enough to convey the concept. Aiming for "rough" and "right" ensures that time is not wasted 19 developing concepts which eventually become "straw men" concepts and do not inform the final design (Thomke 2003). In explaining their prototyping philosophy, IDEO founder David Kelley also notes that following the three R's prevents designers from becoming too attached to the concepts that their prototypes represent. This effect, known as design fixation, has been well-documented and studied (Jansson and Smith 1991; Linsey et al. 2010). Recent studies by Viswanathan and Linsey argue that design fixation can be mitigated by generating rapid prototypes (Viswanathan and Linsey 2013). By mitigating design fixation, designers hope to enable themselves to consider a wider range of available options, encompassing much of Ward's original intent for SBD. 4 2.3.2 Tolerance for Under Defined System Specifications Another aspect of Toyota's seemingly paradoxical efficiency from the Ward et al. (1995) study was its designers' tolerance for under defined system specifications. Toyota managers were intentionally reluctant to lock down system specifications prior to key decision making points throughout the design process. Examples included managers reporting only approximate subsystem specifications to target to their suppliers, and delaying the release of final specifications to these suppliers until late in the design process. A particularly poignant example was in body design, where Toyota postponed fixing body hardpoints in the vehicle architecture. As one of Toyota's general managers put it, "the manager's job is to prevent people from making decisions too quickly." This was in stark contrast to the more traditional design approaches adopted by other Japanese and US automakers, who would try to lock down specifications as soon as possible. In their view, this was the only way to move the design forward in the process. Characteristic 2 in Civil Engineering, Management, and Software Applications Other applications in the literature have also documented an ability to advance the project at hand without having defined all specifications up front, similar to Toyota. Let us consider a 4 As we will find later on in this thesis, the intention of mitigating fixation may not necessarily guarantee Ward's desire for not ruling out options from consideration. We will touch upon this point at the end of this chapter and again throughout the thesis prior to our dedicated discussion on fixation in Chapter 7. 20 case study from civil engineering, specifically in airport design. Gil et al. undertake a study of upstream and downstream tasks in airport design, where upstream represents "base building" tasks that provide the service space for occupancy, whereas downstream entails fit-out subsystems that make the space functional (e.g. check-in baggage counters, baggage screening machines, etc) (Gil, Beckman, and Tommelein 2008). Traditionally, it would seem counterintuitive to delay decisionmaking on upstream tasks while developing downstream tasks. However, this approach is employed effectively to gain total development time between the concurrent development of upstream and downstream tasks. Managers on this project also pointed out the benefits of their "progressive fixity" to decision making, where intentionally delaying certain decisions built greater flexibility into their project. This mitigates their exposure to risk from events such as shifting requirements or availability of materials. Further exploring these concepts on a theoretical basis from the management science literature are Thomke and Reinersten (Thomke and Reinertsen 1998). They note the need for modem day corporations to incorporate agility in their PD processes by adopting "development flexibility." In a series of hardware and software applications in the computer industry, their results advocate "progressively locking down requirements," thus agreeing with the airport design and Toyota examples offered above. Furthermore, they support making "piecewise commitments vs. binary choices," where decisions are made on certain aspects of the project despite not having 100% of project specifications approved in advance. This increases flexibility, driving down the cost of a.potential change down the line, and supporting the idea that Toyota's practices of decision delay may have actually helped promote efficiency. Thomke and Reinersten's study of software applications was apt, as we find significant support for the second Characteristic of SBPD in agile software development. As alluded to earlier, the emphasis of rapid development in Scrum and other methodologies is done so that software can accommodate a highly dynamic environment of rapid technological change, steep competition, and quickly shifting customer requirements (Sutherland et al. 2007). Hence, agile developers postpone decision-making on the final specifications of the UI, especially in an area where customers need to see a final product in order to get a sense of their preferences (Thomke and Reinertsen 1998). 21 Characteristic 2 in Engineering Design and other Domains In the discussion above, we observe that the tolerance of under defined specifications can be interpreted as a tool to manage uncertainty and to mitigate risk. Building upon this thinking, an excellent way to introduce tolerance for under defined specifications and to mitigate risk is to adopt a real options approach to PD (de Neufville 2003). Recognizing this connection, Ford and Sobek use a real options approach to analytically investigate the point beyond which delaying decision making no longer produces practical benefits (Ford and Sobek 2005). Another method to manage uncertainty via this Characteristic, particularly for companies that manage large portfolios of individual products, is to develop a common product platform. Meyer and Lehnerd define a product platform as "a set of common components, modules, or parts from which a stream of derivative products can be efficiently developed and launched" (Meyer and Lehnerd 1997). Robertson and Ulrich put forth a similar definition, with emphasis on asset sharing across a "set of products" (Robertson and Ulrich 1998). From these definitions, the link to set-based thinking has been established, with companies successfully employing product platforms to mitigate uncertainty, such as Black & Decker's universal electric motor (Simpson 2004). Furthermore, Simpson notes that one can "stretch" or "shrink" platforms in order to create specific products capable of targeting market niches. Notably, to create a product from scratch to tailor niches implicitly assumes high risk, where a better strategy would be to utilize the underspecified system that is a product platform. Finally, several in the engineering design community have and continue to mathematically characterize or model scenarios that deal with the uncertainty of specifications that is at the crux of Characteristic 2. Representations that utilize traditional set theory include (Finch and Ward 1997; Malak, Aughenbaugh, and Paredis 2009; Malak and Paredis 2010; Panchal et al. 2007), whereas those that explore employing fuzzy sets include (Antonsson and Otto 1995; Shan and Wang 2004). Regardless of specific methodology, these works give credence to applying sets to analytically model the tolerance for under-defined system specifications. 2.3.3 More Efficient Communication among Subsystems A common practical argument for using set-based methods or techniques is that sets enable fundamentally richer, more efficient communication. Ward et al. (1995) offer a simple thought 22 experiment to illustrate their point. They argue that when a team of individuals is trying to schedule a meeting time, a set-based approach is most efficient. That is, if each participant simultaneously communicates their availability via sets of time, the meeting is scheduled much quicker than if one participant were to check with another if he/she is available at a given time, and so forth. Beyond this thought experiment, Ward et al. found support for their arguments in Toyota's PD processes, where Toyota and its suppliers were found to establish communication with each other less frequently for a shorter total duration of time than their US counterparts employing traditional design methods. For those practicing traditional design methods, constant communication among collocated engineering teams was a given. The inherent richness of set-based communication has been noted by a few authors in the engineering design community. Madhavan, Seepersad, and colleagues have developed a set-based method for multidisciplinary, distributed design optimization problems. On test problems, they find that their set-based method yielded solutions within 10% of the optimum with only 1 global iteration, translating to 90% less computation expense (Carlos et al. 2006; Madhavan 2007; Seepersad, Shahan, and Madhavan 2007). These results are indicative of the shortcomings of traditional point-based approaches, which, void of the rich information of set-based methods, rely heavily on costly iteration to deliver similar results. Other engineering design authors motivated by set-based efficient communication include Malak and colleagues. Building upon previous work (Malak and Paredis 2010), Parker and Malak develop mathematical models which represent the capabilities of ill-characterized systems, which they call technology characterization models (TCMs) (Parker and Malak 2011). These TCMs intend to facilitate communication of subsystem level performance characteristics, allowing for the direct comparison of sets of heterogeneous concepts within the same performance space, similar to traditional Pareto frontiers. Also inspired by Toyota's example has been the construction academic community, with Parrish et al. offering a SBD framework for the delivery of reinforcing bars in cast-in-place concrete (Parrish et al. 2007). They note that the ruling paradigm in the construction industry is a traditional, point-based design (PBD) approach featuring long delays in passing designs to different agents in the design process. These delays arise from the high amount of rework at each step when a new agent in the design process contributes to the overall design. Parrish et al.'s 23 approach instructs all agents in the rebar design process to contribute their input concurrently, accommodating rich sets of information from each agent and producing a much more efficient communication scheme, bypassing the inefficiencies of iteration. The efficiency of this proposed approach has been recognized, leading other authors in the community to further develop Parrish et al.'s framework (Lee, Bae, and Cho 2012; Sacks et al. 2010). Of course, the efficient communication paradigm of SBD and set-based thinking in general can be found in other theoretical constructs in design, such as shape grammars (Stiny 1980). Orsborn et al. describe shape grammars as "production systems created by taking a sample of the whole for which one is trying to write a language" (Orsborn et al. 2006). Shape grammars, complete with a vocabulary (i.e. a set) of shapes and shape rules can represent "the language" of a given product's design. Orsborn et al. use shape grammars as a method to explore the development of crossover SUV body architectures by enabling the designer to produce a multitude of designs that fall with the constraints of the grammar. The intent is to enable the exploration of new and unique designs that may not have been imagined by the designer. Thus, using sets of shapes, designers can efficiently communicate their ideas on vehicle body architecture, while corresponding shape rules help constrain the design space. 2.3.4 Emphasis on Documenting Lessons Learned/Knowledge While discussing Toyota's practices, Ward et al. (1995) note that in general "designers are notoriously resistant to documenting their work, partly because they sense documentation is generally useless." Toyota, however, expects rigorous documentation of design knowledge via "lesson-learned books." Ward et al. provide an example of a Toyota die-designer, who maintains a book of sixty to seventy different ranges of specifications that would ensure the fender design's manufacturability. They report that similar books are maintained for every body part, which, developed over fifteen years at that point, provide detailed knowledge of what potential designs can (and cannot) be implemented. These lessons learned books are noted to promote institutional learning and to ensure that detailed technical knowledge of one's system is not lost over the years-rather, it is accrued. Similar case examples across domains are noted by Thomke (Thomke 2003). For instance, he highlights the example of pharmaceutical companies stockpiling small quantities of discrete 24 chemical compounds in "chemical libraries." These libraries are the result of years of prior drug development projects, maintaining "several hundred thousand compounds and information on their specific properties." In fact, an emphasis on documentation has gained traction in smaller scale firms such as IDEO. Thomke describes IDEO's "Tech Box," which is a "giant 'shoebox' for curiosities and interesting gadgets from prior IDEO projects." Of course, the intent of the chemical libraries and the Tech Box are the same, which is to help foster new PD. While the industry cases above demonstrate how sets of information can be documented to help promote knowledge retention, a few academic case studies have established a direct connection between following SBD or SBCE processes and learning. In a series of case studies by Raudberget in which he introduces companies to SBCE methodologies, he notes that for one of the companies he studied, "the value of the SBCE project was to identify the knowledge gaps in their core technology" (Raudberget 2010). Furthermore, he notes that by "knowing the limits of their technology," the said company could "have an appropriate set of solutions ready for future offers." Similar results were found in Madhavan et al.'s study of applying SBD principles to Schlumberger, where "archiving design knowledge in the form of sets of solutions... can be used in future design activities" (Madhavan et al. 2008). In fact, this idea can be found in other work in design theory, such as function structures. Pahl and Beitz describe the ability to define an overall function that expresses the relationship between inputs and outputs of system independently of the design solution (Pahl and Beitz 1988). They define a function structure as the combination of individual sub-functions that provides a representation of the overall function being analyzed. Pahl and Betiz point out potential efficiencies of employing function structures, noting that "if existing assemblies can be assigned directly as complex sub-functions, the subdivision of the function structure can be discontinued at a fairly high level of complexity." Thus, documenting system knowledge via sub-functions from prior experience in a set that is readily available for recall in the future may help save "a great deal of time and money," similar to how Toyota used their lessons-learned books. Another tool that can enable set-based knowledge representation is affordances. Maier and Fadel view an affordance-based approach as an alternative to function analysis, where affordances describe a potential behavior between two or more subsystems within a larger designer-artifact25 user complex system (Maier and Fadel 2008). The goal is to articulate relationships that are unaccounted for in a function based approach to design, such as non-function based customer requirements (Maier and Fadel 2008; Cormier, Olewnik, and Lewis 2013). Thus, an affordancebased approach, along with function structures, can enable set-based thinking. 2.3.5 Support for Decentralized Leadership Structure and Distributed, Non-Collocated Teams This section is the first of three which explore insights from Toyota's unique organizational structure. Evidence for Characteristic 5 is apparent from examining Toyota's relationships with its suppliers, as reported in Ward et al. (1995). These relationships often provide considerable autonomy to the suppliers, with examples such as radiator and alternator manufacturer Nippondenso (now Denso) offering sets of potential alternators for Toyota to consider and choose from, rather than being tasked to develop an alternator that meets specifications initially determined by Toyota. Therefore, Nippondenso is accorded considerable weight in Toyota's decision making process, as the set of options offered to Toyota has been determined by Nippondenso's estimates of Toyota's needs-alluding to decentralized leadership (where decisions do not necessarily originate from the top). Furthermore, Nippodenso and many of Toyota's suppliers do not operate in-house, thus exhibiting support for distribution and noncollocation of subsystem designers. Decentralized Leadership Structure We begin by exploring support for a decentralized leadership structure in set-based methods. As noted earlier, the Scrum methodology of software development provides excellent examples of SBPD. The Scrum method is carried out by Scrum Teams. There are three core roles on a Scrum Team: Product Owner, Development Team, and Scrum Master (Schwaber and Sutherland 2013). While detailing the exact responsibilities of each role is beyond the discussion of this work (details can be found in (Schwaber and Sutherland 2013)), what is notable about the separation of labor in a Scrum Team is that there is no role which bears the responsibilities of a traditional project manager. For instance, while the Scrum Master "is responsible for ensuring Scrum is understood and enacted" (Schwaber and Sutherland 2013), he/she does not bear personnel management responsibilities that are typical of a project manager in a company with a 26 traditional, centralized leadership structure ("Scrum Primer"). The reluctance to adopt a traditional leadership structure stems from the belief that it would produce sub-optimal results. Exploring the role of varying levels of cooperation in concurrent engineering processes are Lewis and Mistree (Lewis and Mistree 1998). Using a decision-based design perspective, they develop a game theoretic model to represent non-cooperative situations in concurrent engineering processes. They note that the resulting model "is similar in concept to the idea of set-based concurrent engineering.. .explored in (Ward et al., 1995), except there is no communication among the various designers who are generating sets of solutions." Nonetheless, the reason why they observe similarities is due to their model's ability to represent decentralized leadership structures. This game theoretic model would later be further developed by Lewis and colleagues to investigate various distributed design problems (Chanron and Lewis 2005; Devendorf and Lewis 2011; Ghosh, Devendorf, and Lewis 2014). It is noted that this framework works well to model nonhierarchical systems of designers (Devendorf and Lewis 2011)-which supports the decentralized leadership observation. Furthermore, the Nash equilibrium that represents the converged solution in the model is the result of intersecting each designer's rational reaction set, which is the set of solutions that optimizes that individual designer's objective relative to the decisions of other designers. Hence, we observe support for the use of set-based modeling and decentralized leadership, although we note that the models of Lewis and colleagues do not agree with each of the observed characteristics of SBPD as outlined in this work (e.g. Characteristic 3). As we will find in Section 2.4, not all observed processes can be classified strictly as SBD or PBD; rather, they may fall somewhere on a spectrum between the two. Arguments for Distribution, Non-Collocation To test a set-based PD method's support for distributed, non-collocated teams, Sutherland et al. study whether the aggressive development schedule and culture of Scrum can be sustained across a multinational team of collaborators (Sutherland et al. 2007). Given certain features of Scrum, such as frequent Scrum meetings, not many had ventured to explore whether Scrum would be practical beyond the familiarity of small, collocated settings that entailed its predominant usage. Thus, Sutherland et al. perform a case study tracking a distributed team of 56 developers across three countries and witness the most productive Java development project to have been 27 documented at the time of their publication-a testament to set-based PD practices supporting distributed, non-collocated teams. Another application area for set-based PD practices has been the Navy, specifically in the design of the Ship-to-Shore-Connector (SSC) (Mebane et al. 2011). Mebane et al. study the Navy's desire to develop the SSC on an "extremely aggressive" time schedule, leading the Navy to adopt an SBD approach to its design. While drawing conclusions from the study, they note "the Navy's ability to develop a design with a distributed team with much of the design team personnel remotely located in the field was certainly proven," demonstrating sources of support for this Characteristic of SBPD beyond just the private sector. 2.3.6 Supplier/Subsystem Exploration of Optimality An interesting feature of a decentralized leadership structure that often provides subsystems with greater autonomy in the design process is that it encourages suppliers and subsystems to take initiative in exploring optimality. Continuing with the aforementioned example of Nippondenso, Ward et al. (1995) describe its approach to look for radical performance breakthroughs-rather than incremental improvement-in their designs, helping ensure their relevance in the market of automotive suppliers. Hence, Toyota benefits from having a supplier that prioritizes being at the forefront of its technology, helping propel Toyota's search for globally optimal designs. Other examples in design practice which feature this Characteristic include Madhavan et al.'s previously discussed study (Madhavan et al. 2008). The authors help substantiate this claim by pointing out how their set-based method encouraged Schlumberger to conduct tradeoff analyses at the subsystem level, such as the tradeoff between mass and stiffness in a given chassis design. Studying design spaces across interfaces was not prioritized prior to applying set-based methods, implying a less formal exploration of optimality when compared to applying SBD. Examining the Navy's SSC project once again, Mebane et al. reported that Systems Engineering Managers would communicate "ranges of solutions with associated derived requirements for various systems.. .rather than develop a single point solution" (Mebane et al. 2011). The result was the consideration of a "far greater range of options.. .than would have been 28 considered in traditional point-based design evolutions...in a much shorter time period." Of course, this extended exploration is a key feature towards the exploration of more optimal solutions. Finally, beyond case study examples, the aforementioned example of technology characterization models (TCMs) illustrates this Characteristic well. As noted earlier, TCMs were developed to address the problem of how to present and explore the technical capabilities of certain types of ill-characterized complex systems (Parker and Malak 2011). Algorithms developed from this work feature a bottom-up flow of information (Galvan and Malak 2012). The rationale for this approach is that when subsystems explore optimality, this should scale up to the greater system level in efforts to approach a global optimum. In essence, these efforts by Malak and colleagues are to mathematically model an approach that resembles the flow of information that exists between Toyota and its suppliers, such as Nippondenso. 2.3.7 Supports Flow-up Knowledge Creation As alluded to in the discussion of the previous section, the supplier/subsystem exploration of optimality lends itself to supporting flow-up knowledge creation. What is truly indicative of this Characteristic is the timing of suppliers' presentations to Toyota. Delivered about 36 months prior to the start of new model production, these presentations allow suppliers to describe their latest developments to Toyota, which includes prototypes, large amounts of test data, and related information. The critical difference between Toyota and competitors as reported in Ward et al. (1995) is that while US companies would develop specifications in-house prior to these types of presentations, Toyota would develop its specifications almost two years after its supplier presentations. In other words, it would take information from these presentations to guide critical decision making behind setting final specifications, indicating a bottom-up flow of information, as opposed to the traditional top-down flow where specifications were handed down by US manufacturers. An excellent example of how SBD was applied for its flow-up qualities is provided by the aforementioned study by Parrish et al. (Parrish et al. 2007). As noted earlier, the intent of Parrish's work is to involve all agents across the construction supply chain (e.g. owners, architects, structural engineers, concrete suppliers, etc.) in the design decision-making process. This is in contrast to a 29 traditional model where a structural engineer will make all decisions up front, leading to a single, often suboptimal, design. This flow of decision making is decidedly top-down, as opposed to Parrish's approach which empowers subsystems (i.e. agents such as concrete suppliers) in the decision-making process, leading to a bottom-up flow of information. Academic examples of this Characteristic include the work of Malak and colleagues as noted in Section 2.3.6 regarding TCMs. By optimizing subsystems first, the flow of information is certainly bottom-up. Other examples include work from Shan and Wang, who develop an algorithm using rough sets to map "the performance space to the design space" that "could be potentially used to explore and visualize the entire design space" (Shan and Wang 2004). The performance space that these authors refer to is attributes at the subsystem level, whereas the design space is the greater system level. Thus, similar to Malak and colleagues, these authors wish to use sets to map performance capabilities at the subsystem level to a higher level system level space to pursue globally optimal design-entailing a flow-up approach. Finally, more recent work is presented by Canbaz et al., who in studying distributed design processes, expand the "bottomup design approach with exploring agent-based modeling techniques into the set-based design concept" (Canbaz, Yannou, and Yvars 2013). 2.4 The Two Principles of Set-Based Thinking Having reviewed evidence for the 7 Characteristics of SBPD, we now turn our attention to the development of 2 Principles of Set-Based Thinking (the 'Principles'). This section is organized as follows. In Section 2.4.1, we present the Principles, and in Section 2.4.2 we introduce the SetBased/Point-Based Process Spectrum, which enables us to analyze the degree to which certain processes exhibit set-based or point-based characteristics. Finally, in Section 2.4.3, we detail how the 7 Characteristics of SBPD lead us to obtaining 2 Principles of Set-Based Thinking. 2.4.1 The Two Principles of Set-Based Thinking At this point in our study, we make the following observations: - SBD is not formally defined, yet numerous authors have studied its process inspired by the example of Toyota 30 * The 7 Characteristics of SBPD are a filter which allow us to review work that exhibits setbased practice A major challenge in our task to define what SBD means is determining what work to include and what not to include in our review-which is the function of the 7 Characteristics. Having reviewed the literature of set-based practice in Section 2.3, our task now is to find the most common themes and ideas in this reviewed work that capture SBD, which we term the Principles. Our review supports the finding of two, independent Principles. These Principles are regularly woven into the ideas discussed in the work reviewed, both explicitly and implicitly. At this phase of our preliminary study, we cannot guarantee that these are the only two Principles; however, our goal was to distill the network of complex ideas in Section 2.3 into the most simple, fundamental representation possible, which led to the articulation of these two Principles. As a preliminary validation of this approach, in Section 2.4.3 we explain how the 7 Characteristics of SBPD directly embody the Principles. But first, we present the Principles: 1. Considering sets of distinct alternatives concurrently 2. Delaying convergent decision making Central to the ideas of SBD and related work is the concept of concurrent development, which is captured by Principle 1. Concurrent development describes the simultaneous development of distinct concepts. This is in contrast to traditional, point-based processes, where designers establish commitment early on to one concept which receives the singular focus of the designers involved. It is important, however, to establish the distinction between the development of sets of distinct concepts and the development of sets of variants of a single concept. The former captures the true ethos of SBD as noted in the introduction, which is to consider as many unique options as possible. The latter exhibits a limiting case of SBD and may regress to traditional, PBD processes. To demonstrate the difference between sets of distinct concepts and sets of variants of a single concept, consider finding a solution to travel from the city of Rochester, NY, USA to Toronto, ON, Canada. One may consider renting a car, taking a bus, or hiring a limousine service. However, these options entail variants of a single concept: to travel by highway. Other potential 31 options include train, airplane, and even a ferry ("HSC Dolphin Jet" 2013). While this is a simple example, it illustrates that comparing sets of variants of a concept limits the solution space explored and does not provide the full intent of Principle 1. Nuances such as these will be important to keep in mind when analyzing the spectrum of set-based and point-based processes, as presented in Section 2.4.2. Equally important to SBPD is the second Principle: delaying convergent decision making. Note that this Principle does not advocate delaying decisions for the sake of delaying decisions; rather, it supports delaying convergence to a single solution. Some authors highlight this distinction by pointing out that decisions should be delayed until the "last responsible moment" (Gil, Beckman, and Tommelein 2008). The emphasis on "responsible" highlights how decision delay as a strategy should not be exercised as an excuse to simply not make decisions. Instead, it should facilitate the consideration of more options. Specific to SBD, following this Principle results in a process where the solution space gradually narrows to a single solution, as opposed to traditional, point-based representations, which promote rapid convergence and reiteration on a single concept that may require rework calling for re-expanding the solution space down the road (Sobek, Ward, and Liker 1999), which intuitively is a source of inefficiency. 2.4.2 The Set-Based/Point-Based Process Spectrum In previous sections, we alluded to the idea that PD processes should be qualified on a spectrum that encompasses SBD and traditional, PBD processes. Such a spectrum can help highlight how some processes may exhibit certain characteristics of set-based PD but still not represent an ideal case of SBD. A question that has gone unaddressed in the body of SBD research is finding heuristics for determining when it is appropriate to apply set-based vs. point-based techniques. Thus, a spectrum that helps us classify to what degree a certain process represents a complete implementation of SBD is a first step towards answering that question. In order to develop a set-based/point-based process spectrum, we first need a definition for PBD. This is a non-trivial task and is similar to attempting to define a primitive notion-in the sense that point-based, traditional design processes encompass that with which we are already familiar. In his discussion of agile product development processes, Lansiti characterizes traditional design processes as those where detailed design and implementation occurs only after 32 concept selection occurs (lansiti 1995). Along these lines, Ulrich and Eppinger, commenting on a generic (i.e. traditional) product development process, posit the following about concept development, "alternative product concepts are generated and evaluated, and one or more concepts are selected for further development" (Ulrich and Eppinger 2012). While these definitions capture some of the distinctions of point-based design relative to SBD, they are generally incomplete and refer only to the interface between the concept development and detailed design phases. This is limiting since the Principles can manifest themselves at all phases across the design process. Therefore, we propose a working definition for point-based and traditional design processes that is defined relative to our definition of SBD. Similar to how temperature is defined on a relative basis to a given reference point (absolute zero), we wish to define point-based and traditional design relative to the 2 Principles of Set-Based Thinking. Let A = Principle 1 ("considering sets of distinct alternatives concurrently") and B = Principle 2 ("delaying convergent decision making). From these statements, we can define SBD succinctly: SBD EA A B In other words, SBD (or more generally, SBPD) is defined by Principle 1 and Principle 2. We define traditional design relative to SBD in the following way: TraditionalDesign -,SBD = -,(A A B) = -,A V -,B In other words, traditional design is simply "not SBD." Thus, by DeMorgan's Law, it encompasses a lack of Principle 1 or a lack of Principle 2, or a lack of both Principles 1 and 2, since we use the inclusive definition of "or" in the notation above. In fact, where both a lack of Principle 1 and a lack of Principle 2 are exhibited, we have a special case of traditional design, which is pure PBD. The rationale for this working definition is that if a process exhibits one of Principle 1 or Principle 2, then it inherently bears some set-based characteristics, and thus should fall in a spectrum between the poles of SBD and PBD, where pure PBD represents cases that do not bear any set-based characteristics. This spectrum is demonstrated visually in Figure 1. 33 SBD PBD BOTH: Either/or Neither: - Principle 1 - Principle 2 - Principle 1 - Principle 2 - Principle 1 - Principle 2 Y11 SBPD -"YJ Traditional Product Development Figure 1: The Set-Based/Point-Based Process Spectrum An important question to ask is whether the Principles can be considered independently in the proposed spectrum. That is, can Principle 1 act independently of Principle 2, and vice-versa? For instance, does a process exist where we consider sets of alternatives concurrently without delaying convergent decision making? The answer is yes-consider the example of a morphological matrix. This method hastens the consideration of a very large number of distinct alternatives (Pahl and Beitz 1988). Still, Pahl and Beitz note the "weakness of this approach" by noting "the very great, theoretically admissible but practically unattainable, number of solutions must be reduced at the earliest possible moment." Thus, using a morphological matrix is not conducive to promoting decision delay. Now let us consider the opposite case: a process that delays convergent decision making but that does not necessarily promote considering sets of distinct alternatives concurrently. A good example here is the overuse of a product platform, such as Chrysler's K-car platform in the late 1980s (Simpson 2004). While product platforms clearly enable the delay of convergent decision making (product can be tailored later to target niches, see Section 2.3.2), designers in a push for leveraging commonality may feel inclined to use it more often than they should, thus limiting the consideration of distinct alternatives. Thus, we have demonstrated via example how the two Principles can act independently of one another, and critically how they must both be present in order to facilitate true SBD. 34 2.4.3 Deriving the Two Principles from the Seven Characteristics of Set-Based Product Development In this section, we detail the process by which the 7 Characteristics of SBPD reveal the 2 Principles of Set-Based Thinking. Note that some of the 7 Characteristics support one Principle more than the other. However, considering the 7 Characteristics holistically allows us to arrive at developing the 2 Principles. Thus, we go through each Characteristic demonstrating their collective support for the Principles. Emphasis on Frequent Lo-Fidelity Prototyping Prototypes enable designers to compare several distinct alternatives as embodied by the aforementioned design language argument in (Yang 2005), thus exhibiting support for Principle 1. Furthermore, the emphasis on prototyping early and with lo-fidelity (Haggman, Honda, and Yang 2013) enables further concurrent development, thus postponing convergence to a single concept and supporting Principle 2. Tolerance for Under Defined System Specifications Designers using set-based techniques are comfortable with incomplete information, and thus can work with concepts without necessarily forcing convergence to continue development, as demonstrated in (A. Ward et al. 1995; Gil, Beckman, and Tommelein 2008; lansiti 1995; Thomke and Reinertsen 1998; Mebane et al. 2011)-thus supporting Principle 2. In fact, incomplete information is often fostered by system managers (A. Ward et al. 1995; Sutherland et al. 2007; Gil, Beckman, and Tommelein 2008; Schwaber and Sutherland 2013), inducing designers to further explore the solution space and reinforcing concurrent development and support for Principle 1. More Efficient Communication among Subsystems Communicating via sets is inherently more efficient than passing information sequentially and iterating on a point-based solution (A. C. Ward and Seering 1989; A. Ward et al. 1995; Parrish et al. 2007; Madhavan et al. 2008; Mebane et al. 2011; Parker and Malak 2011), supporting concurrent development and Principle 1. 35 Emphasis on Documenting Lessons Learned/Knowledge Several examples described in Section 2.3.4, from Toyota's lessons learned books, pharmaceutical companies' chemical libraries, and IDEO's Tech Box illustrate how documenting knowledge can lead to the consideration of a wider set of options, supporting Principle 1 [15]. In addition, by allowing designers to reason about why options should remain in play such as with function structures (Pahl and Beitz 1988), Principle 2 is upheld. Support for Decentralized Leadership Structure and Distributed, NonCollocated Teams The concurrent development of distinct alternatives is easily performed in distributed, noncollocated settings (A. Ward et al. 1995; Sutherland et al. 2007; Mebane et al. 2011), unlike traditional processes that rely on iteration and require centralization. Thus Principle 1 is sustained. Supplier/Subsystem Exploration of Optimality System managers of set-based processes will not provide predetermined specifications to suppliers (A. Ward et al. 1995; Malak and Paredis 2010; Parker and Malak 2011), who in their search for optimality will need to adopt the concurrent development of distinct alternatives (Parrish et al. 2007; Madhavan et al. 2008), fostering Principle 1. By tasking subsystems to explore optimality, the system manager also inherently delays convergence by fostering set-based exploration as opposed to point-based exploration (A. Ward et al. 1995; Raudberget 2010; Madhavan et al. 2008), supporting Principle 2. Supports Flow-up Knowledge Creation Following from the last Characteristic, supplier exploration of optimality, which is achieved via concurrent development, fundamentally enables a flow-up knowledge creation, thus promoting Principle 1. Similarly, from Nippondenso's example we saw how system managers (i.e. Toyota) delay convergent decision making until subsystems perform their tasks of exploring subsystem level optimality, enabling a bottom-up flow of knowledge that is conducive to supporting Principle 2. 36 2.5 Case Example: Set-Based Thinking in Function Structures Having developed the Principles in the previous section, we now turn to a brief case example to illustrate how these Principles manifest themselves in a common theoretical construct in design: function structures. We intentionally choose to explore function structures as they are an example of work that is not nominally related to SBD but that we review in this work as it & features set-based reasoning. An example of a real-world function structure for the Black Decker@ VersaPakTM Cordless Screwdriver (Dahmus, Gonzalez-Zugasti, and Otto 2000) is provided in Figure 2. Further reading on the background and theory of function structures is referenced in Section 2.3.4. In a function structure, each sub-function can embody a wide array of physical configurations-the function structure does not discriminate in this regard. Refer to the example in Figure 2. While the sub-function "input signal" has been implemented via a sliding switch in the final product shown, the sub-function itself enables consideration of options such as a pushbutton switch or a rotary switch to name a few examples. Thus, function structures clearly enable the consideration of sets of distinct alternatives, or Principle 1. However, the real value of function structures in the context of set-based reasoning is realized by plotting the flow of signals and other system interactions, nicely articulating constraints on the solution space in which designers explore. This allows designers to reason about which alternatives to explore and which to not explore-a line of reasoning that extends back to the genesis of SBD (A. C. Ward 1989; A. C. Ward and Seering 1989). In this sense, it facilitates Principle 2, the delay of convergent decision making, because all physical instantiations that meet the requirements of the function structure are carried through the design process. This is a methodical way to practice decision delay, as opposed to morphological matrices (see Section 2.4.2), where coarse heuristics must be applied in order to rule out options that do not meet system physical constraints. Such heuristics may eliminate more options than required, thus preventing the delay of convergent decision making. 37 Hand force -~------~ Register Battery --~ Battery Unregister ~-~ Battery Battery Transmit Electricity ~--------+----------~-- Force into opposite hand Noise. Heat - ----j Torque Prevent Back Rotation Screw Turn Screw Permit Bit Positionin Hand Force Hot turning screw _____ _..,.. Bit position Reaction force Bit L~ Bit --~~~~~secur ~e~---~~~~~~~--~~-- Fm:eruo · opposite hand Figure 2: Function Structure for a VersaPak TM Cordless Screwdriver. From (Dahmus, Gonzalez-Zugasti, and Otto 2000). 2.6 Discussion and Conclusion In 1995, Ward and colleagues documented Toyota' s unique PD practices as examples of SBCE, generating a flurry of research into SBD and set-based reasoning across academic communities, such as engineering design, management science, civil and structural engineering, software design, and ship building, to name a few examples. Over this time period, SBD has never been formally defined, despite many authors having studied its process inspired by the example of Toyota. In this chapter, our objective is to inductively identify common themes and ideas from the body of work that has emerged since Ward et al. (1995)'s original study. From their study, we identify 7 Characteristics of Set-Based Product Development, which essentially are observed attributes of set-based processes. We then employ these Characteristics as a filter by which to choose what research to review in this work. We review work that is both nominally tied to SBD and also that which actively features set-based reasoning. After studying the sample of literature 38 that we generate, we find two, independent principles that are both explicitly and implicitly integrated into the ideas of the literature reviewed in this work. These are 2 Principles of SetBased Thinking, which are 1) considering sets of distinct alternatives concurrently, and 2) delaying convergent decision making. We use these Principles as a basis by which to offer a working definition for SBD. We also introduce the Set-Based/Point-Based Process Spectrum, which provides a preliminary way to represent the idea that some processes may exhibit certain characteristics of Set-Based Thinking without being an ideal case of SBD, thus presenting a hybrid of set-based and point-based reasoning. Taking a step back, we may ask: what do the Principles of Set-Based Thinking mean in practice? What does it mean for a process or a tool to be classified on the "SBD" end of the SetBased/Point-Based Process Spectrum? Considering the simultaneous effects of concurrent development and delaying convergent decision making, we can assert that the consequence of SetBased Thinking is the consideration of more options. This result should not be surprising, as we note that the motivating ethos of SBD was to create a framework in which options would not be eliminated from consideration unless there was a logical reason to do so. In a sense, the intellectual foundation of SBD makes the normative assumption that considering more options throughout the process is inherently a good practice in design and PD. Considering more options as a heuristic is the goal of many creativity and innovation techniques explored in the literature and is promoted in broad-based paradigms such as Design Thinking (Dym et al. 2005). In making the case for SBD, many practitioners have touted the promise of better, novel outcomes to result from having considered more options (Singer, Doerry, and Buckley 2009; Mebane et al. 2011). However, in practice, many of these practitioners noted a lack of truly novel concepts to result from following the process, reflecting that the end outcomes obtained were not surprising and could have been anticipated ("Ship-to-Shore Connector Implementation of Set-Based Design" 2009). This observation raises a host of questions. First, is it valid to assume that considering more options throughout the process will generate better outcomes? What makes considering more options difficult? What are the ensuing implications for Set-Based Thinking and the way we think about the process in general, especially in the front end when most divergent thinking occurs? 39 These questions and others shall form the underlying intellectual core to our development of a novel framework for product development, which shall be the focus of the remainder of this work. 40 3. A New Framework for the Product Development Process Having grounded ourselves in the literature in the previous chapter, this chapter begins to lay the groundwork for the development of a novel framework for PD, which shall be the focus of the remainder of this work. framework. In Section 3.1, I identify the motivation for developing such a Following this, I introduce the terms which shall define the language of this framework in Section 3.2, with the roles each term plays explained in Section 3.3. In Section 3.4, I reference prior use of these terms in the literature, and I conclude the chapter in Section 3.5 by identifying some of the ideas which shall be explored in further detail in the chapters to follow. 3.1 Motivation and Background In advancing the inquiry on SBD and related techniques that we began in Chapter 2, we were troubled by an observation that became abundantly clear not only as the literature review progressed, but also during the formal presentation of this work at the 2014 ASME International Design Engineering Technical Conferences (Ghosh and Seering 2014). That is, despite the welldocumented interest in SBD as a topic of study, there existed a lack of agreement on the use of basic terms and concepts discussed in this literature. Given some of the misunderstanding that ensued during discussion of fundamental elements of the PD process that accompanied SBD, it became clear that this issue transcended the SBD literature and was indicative of a larger issue in the greater design and PD literature. Of course, it would be naYve to assume that we were the first to notice this issue in the design literature. In her reflections of the Human Behavior in Design Conference in 2003, Blessing succinctly describes the problem we face (Blessing 2003). An excerpt of these reflections is provided below: "Although the participants formed a specialized part of the design research community as a whole, the issue of differences in terminology became very clear. Terms borrowed from other disciplines had been interpreted in different ways, the disciplines represented by the participants had their own 41 terminology unfamiliar to the others and the same terms coming from participants from different cultures had a different meaning. This resulted in confusion, discussion and a clearly expressed need to have a common terminology. A common terminology, was seen as essential for a research area. Only in this way a common understanding and building on existing work can be realized." Along these lines, Blessing also notes calls during the meeting for a "common model" for design: "The wish for a common model or a set of (partially shared) models came up frequently. models exsit, most research results remain unconnected. Few Such a model would indicate a better understanding of design and would provide a shared understanding, a basis on which to do research." At the time of this writing, it has been over a decade since Blessing's calls for a common terminology and common model have been published. However, during our experience of articulating the SBD and affiliated literatures, it is clear that the issues underlying Blessing's observations continue to trouble our academic community. As Blessing notes, existing work in our community cannot build on itself without a common understanding of the terminologies and models we use, as misinterpretations of each other's work will not foster an environment conducive to generating the deeper insights about the PD process that we desire. The remainder of this work is my humble attempt to address Blessing's concerns (and in a broader context, Simon's call for the formalization of design science as established in Chapter 1) by developing a novel framework for PD theory and methodology. The following sections will introduce the terminology upon which we shall build this emergent framework, which shall result in a fundamental reinterpretation of certain elements of the PD process. This framework is certainly not presented as the "common model" by which all other work in our community shall be interpreted; rather, it is presented as another way to interpret the basic PD process with which we are familiar. It is ultimately up to the community to determine the usefulness of this framework in addressing the latent research challenges in the PD literature following the publication of this work. In developing this framework, we shall remain cognizant of the challenges we identified at the conclusion of Chapter 2, such as: why is the consideration of more options difficult in practice? 42 And why does considering more options not necessarily guarantee better outcomes? However, before we can address these questions more directly, we must define the lexicon upon which our framework shall develop. 3.2 Definitions In this section, I will outline definitions for the framework. Each subsection is titled with a specific term, followed by the formal definition of that term listed in italics below the subheading. Below the definitions are brief discussions on the respective terms, including examples of the terms in use. For clarity, the first few references to each term (including those referenced prior to their formal definition) are recognized in bold text. Architecture 'An expression ofform of unspecifiedresolution."5 The term architecture is one of the most often used yet ill-defined terms in the design theory literature, as we will note in the subsequent sections discussing the terms of the framework in use. We break this definition into two parts and illustrate each with an example to understand its implications. In this definition, "unspecified resolution" plays an important role as it enables the broad representation of ideas using the term architecture (please see definition below for resolution). For instance, the concept of a V6 engine (before any parameters such as displacement have been formally specified) qualifies as an architecture, as does the fully-specified historical example of the 1950 Lancia Aurelia 1.8L "1800" V6 engine. 6 5 In Section 3.4, this definition of architecture shall be contrasted with prevailing norms for the term in the greater engineering design literature. Before proceeding, to preemptively avoid any misconceptions, it is important for the reader to note that my definition for architecture does not include notions of function (it is only defined in terms of form). This will be discussed in further detail in the following sections. 6 The 1950 Lancia Aurelia was the first production car to feature a V6 engine, where the "1800" engine was the first V6 configuration Lancia offered. 43 The first half of the definition, "an expression of form," can support a variety of interpretations. Given the traditional context of studying problems pertaining to mechanical design within the engineering design community, form is typically thought to imply or represent a physical artifact. Accordingly, we are often conditioned to think of architectures as physical artifacts. However, form may also represent itself in analogues in other domains. For instance, in the context of software, consider a program to aid in the task of prototyping. Here, an architecture may represent a CAD program. In service design, consider the example of designing a way to administer health insurance. In this context, an architecture may represent a health maintenance organization (HMO). As the distinction among product, software, and service design continues to blur now and in the years ahead, a more global definition of architecture is necessary for the future of design and product development research. Many commentators today actively employ this holistic view of PD, with one of the most popular editorials to be distributed within PD and innovation circles the year this thesis was published arguing that companies should be more concerned with controlling "customer interfaces" rather than the traditional supply chains that deliver goods and services (Goodwin 2015). While it remains to be seen whether this argument will be valid in the years ahead, it has captured the interest of those outside our academic and professional community (Friedman 2015). As a research community, we must remain cognizant of popular views of PD practice (in this case the increasing obscurity of the boundaries between physical and non-physical PD) in order for our work to better serve and educate those outside our sometimes insular academic circles. Resolution "The degree to which an architectureis embodied or detailed." The resolution of an architecture is defined in relative terms. In the example provided above with the definition of architecture, the unspecified concept of a V6 engine would be qualify as a "low" or "coarse" resolution architecture relative to the fully-specified 1950 Lancia Aurelia "1800" V6 engine. Another way of saying this would be to call the 1950 Lancia Aurelia "1800" V6 engine a "high" or "fine" resolution architecture relative to the unembodied concept of a V6 engine. 44 As noted earlier, an architecture may simply represent a concept that has not necessarily been expressed physically yet (e.g. via physical prototype, etc.). Demonstrating the relative nature of resolution, when considering engine configurations, other architectures that may be considered to be at the same level of resolution as the unembodied concept of a V6 engine would be those of a V8, or an inline-four layout. Design Parameter 'An attributedescribing a characteristicof form or architecture. Can be describedsuccinctly as an axis of the architectural domain." With this framework, we hope to be able to articulate the progression of ideas throughout the design process. In order to do this, we must be precise about how we are measuring ideas. Essentially, any attribute we use to describe form, either qualitative or quantitative, can be thought of as a design parameter. Examples of design parameters include dimension, mass, material used, and color, to name a few instances. Specification 'An attributemeasuringperformance ofa design. "7 While design parameters describe attributes of architecture (thus helping us articulate characteristics of form), we also need parameters which measure the performance of a concept, idea, or design. This is the role of a specification. Consider the example of a #2 pencil. The #2 (or HB) grading represents an attribute of architecture and thus would be classified as a design parameter. However, attributes such as 1) the bending moment exerted to snap the #2 pencil in half (measured in Newton-meters), and 2) the time (in seconds) and 3) the exerted force required (in Newtons) to erase a standardized sentence written by the pencil all represent specifications. 7 Given the prevalence of the Ulrich and Eppinger's paradigm to describe specifications (Ulrich and Eppinger 2012) in the PD process (as we will address in Section 3.4), I should preemptively clarify that this definition for specification does not include notions of performance targets. In this framework, a specification is essentially a performance variable-it does not however imply target values for such performance. 45 Architectural Domain "The M-space of design parameters which define a design's architecture, where Mis an arbitrary positive integer." The architectural domain is the first of two domains that are central to describing the PD process using the framework. It is the space in which designers can articulate form, using qualitative and quantitative design parameters as descriptive attributes. While the architectural domain itself does not exist in any meaningful tangible form (it is an abstract solution space), an interesting analogy to consider here is a detailed engineering drawing. The information portrayed on such a drawing can be thought of as a subset of the design parameters in the architectural domain, such as dimensions, material choice, and perhaps even information on color choice. Note, however, that not all pertinent attributes depicted on an engineering drawing qualify as potential axes of the architectural domain. For instance, indication of which manufacturing method to use would better qualify as a constraint that acts upon the architectural domain. 8 See Figure 3 for an example. Finally, regarding the nomenclature of this domain, the reader may be wondering why it is not simply called the "design parameter" domain, given that this space is essentially embodied by the design parameters of a particular design. The reason is to emphasize that architectures reside natively in this domain (unlike the specification domain, upon which they must be projected). This distinction is discussed in further detail in the sections and chapters to follow. 8 Since manufacturing method employed does not intrinsically describe form (rather, it acts upon form by for instance influencing its shape, thus influencing the values that certain design parameters take), it is better described as a constraint that acts upon the architectural domain. 46 S . .. .375 3X .37 0 1900 5 -03.00 2X .500 12 .T2MWT.132 SEE 4 NOTE R .375 0 2,100 R 150 200 ~TE 125 2.900 R .100 500 R.2 SECTIO0N 'A-A' .150 -NDTE 1. LET TERING TO BE MACHINED WITH A 60 DEG, ENRAVINO TO(L TO A DEPTH OF .007' DEEP. Figure 3: An Engineering Drawing. The drawing displays detailed information on dimensioning of form in addition to a note regarding a preferred manufacturing method. This is a constraint on the architecture, since the desired geometry can be produced using methods other than machining with an engraving tool, illustrating the point earlier that the manufacturing method does not intrinsically describe form. Image from (Gossard 2009). Specification Domain "The N-space of product specifications, N is an arbitrarypositive integer. Articulates product function andperformance." Whereas the architectural domain exists to describe a product's form, the specification domain exists to articulate its specifications. In the aggregate, it is the performance space in which designers measure project success and whether or not they are meeting the objectives for 47 their design. Accordingly, it is helpful to think of the specification domain as that in which optimization problems in the engineering design literature are solved. Comparisons to existing work on multidisciplinary design optimization (MDO) will be drawn in later sections. Instance 'A specific instantiationofa given architecture. A group ofinstancesmay be sampled to represent the idea behindan architecture." In order to truly embody the ethos of considering more options, one must have a process to compare several distinct architectures which span large parts of the specification domain. To do so, we must develop a sense of what are distinct architectures to compare and what are simply variants 9 of one architecture. The term instance helps us articulate one particular occurrence of an architecture. Furthermore, as implied from the definition, specific instances are often necessary to understand an architecture and its characteristics and behavior. To illustrate instance and groups of instances, consider the BMW 3 Series, produced from 2004 to 2012. At an arbitrary level of resolution, an instance of this particular 3 Series platform is the E90 Sedan (or Saloon). Furthermore, a group of instances at the same level of resolution include the E90 Sedan, the E91 Wagon (Estate), the E92 Coupe, and the E93 Coupe/Cabriolet. At a finer level of resolution, an instance may include the "pre-facelift" 2004-2008 E90 3 Series, shown in Figure 4. An instance of finer yet resolution would be the 2008 328i Sedan (produced for the North American market). 9To understand the notion of a variant, please refer to the Rochester-Toronto travel example provided in Section 2.4.1. 48 Figure 4: A Comparison of the "Pre-facelift" 2004-2008 E90 3 Series Sedan to the Facelifted Sedan Produced for Model Years 2009-2012. The "pre-facelift" sedan is shown on the right for the front and rear views of the sedan.10 Point 'An instance of a given architecture(see instance). Can be used in contrastto "set" to illustrate the final, production version of the design chosen from the initialsolution space (expressed in the architectureor specification domain)." 10 Images from ("BMW 3 Series (E90) 2005-2008" 2015) and ("BMW 3 Series (E90) 2008-2011" 2015). 49 While a point nominally qualifies as an instance, it can be used in a more descriptive sense to illustrate the final production version of a product. This distinction will become clear when describing the process holistically using the framework in Chapter 4. Accordingly, to extend the example provided for instance, a point would be the 2008 328i Sedan, embodied down to all features that were installed at the factory (e.g. a "cold weather package" that includes optional allweather mats, etc.). Set "The collection ofall feasible instancesat a given level ofresolution of a design. Can be used to describe the solution space under consideration." Use of the term "point" bears more meaning in the framework when understood in contrast to "set." This definition for set is consistent with Ward's original notion of "set" (discussed in Section 3.4) which refered to the collective solution space being considered at any point in time, gradually narrowing as the process progresses. Suppose one is given the objective to produce visible light. The set would be to produce electromagnetic radiation nominally between the wavelengths of 400 and 700 nanometers. All wavelengths outside of this spectrum (e.g. gamma rays, far infrared rays, etc.) are not part of the set as they are not feasible instances that meet the given objective. Constraint 'A condition that an architecturemust satisfy. May be representedanalyticallyin the architectural and/orspecificationdomain." As implied from the definition, constraints act upon the solution space (whether examining the architectural or specification domain) by influencing the boundaries of the set under consideration. Certain constraints reveal themselves as the process progresses through time, leading us to classify constraints as either "active" or "dormant," according to the definitions below: Active constraint: A constraintthat is currentlyacting upon the set Dormant constraint:A constraintthatis currentlynot acting upon the set 50 ................ Examples of active constraints include laws of physics (which are active from the onset as one cannot generate feasible options which violate these basic principles) and environmental regulations (for instance, EPA emissions regulations that are in play during the consumer end-use of a motorcycle). Conversely, dormant constraints may include supply chain conditions (prior to the detail design phase of Ulrich and Eppinger's generic PD process (Ulrich and Eppinger 2012)) and anticipated government regulations (e.g. during concept development, where a PD team may be producing a design to be compliant with regulations that are due to become enforced once the product hits the market). One important caveat to note is that in practice, constraints are not activated in a binary sense. That is, one cannot simply classify a constraint as being simply "dormant" or "active;" rather, most constraints undergo a process of being gradually activated as the design process moves forward. This will be alluded to in the following chapters and discussed in further detail in Chapter 8. Feasibility 'A state describing whether or not an architectureis attainable. Expressed analytically as the solution space that is bounded by all active constraints." Feasibility is the condition that explains whether or not to include certain instances within the set. Note that this is a more restrictive definition than notions of technical feasibility, since an active constraint unrelated to technical performance may limit the set that is to be considered, thus making certain options unattainable. Consider the example of desktop 3D printers for personal use. While the additive manufacturing techniques behind these machines have existed for years (e.g. Fused Deposition Modeling, etc.), one of the biggest factors prohibiting the early development of personal 3D printers was that they were cost-prohibitive due to patents on these techniques. Given the active constraint of the cost that consumers were willing to personally pay, the open-source development that one sees in the personal 3D printer sector today was rendered infeasible (Bilton 2015). 51 Table 1: Definitions Summary with Examples Term Definition Example Architecture An expression of form of unspecified resolution. V6 engine, CAD software, HMO (health insurance provider) Resolution The degree to which an architecture is embodied or detailed. Unembodied V6 concept vs. 1950 Lnci 0 ene Design Parameter An attribute describing a characteristic of form or architecture. Can be described succinctly as an axis of the architectural domain (see definition below). Dimension, mass, material, color Specification An attribute measuring performance of a design. For a #2 (HB) pencil: bending moment to break in half, time/force required to erase a standardized sentence Architectural Domain The M-space of design parameters which define a design's architecture, where Mis an arbitrary positive integer. A subset of the parameters included in an engineering drawing Specification Domain The N-space of product specifications, Nis an arbitrary positive integer. Articulates product function and performance. Domain for optimization Instance A specific instantiation of a given architecture. A group of instances may be sampled to represent the idea behind an architecture. 2008 BMW 328i Sedan (instance of the E90 3 Series) An instance of a given architecture (see instance). Can be used in Production version of 2008 BMW 328i contrast to "set" to illustrate the final, production version of the design chosen from the initial solution space (expressed in the architecture or specification domain). Sedan (e.g. with "cold weather package") Point 52 Lancia " 1800" engine Term Constraint Feasibility Definition Example The collection of all feasible instances at a given level of resolution of a design. Can be used to describe the solution space under consideration. Visible light: wavelengths of 400 to 700 nm A condition that an architecture must satisfy. May be represented analytically in the architectural and/or specification domain. Active currently acting upon the set Dormant currently not acting upon the set Active laws of physics, environmental regulations (during consumer end-use) Dormant supply chain (prior to detail design), anticipated government regulations A state describing whether or not an architecture is attainable. Expressed analytically as the solution space that is bounded by all active constraints.upon Develpent ofprson patent expiration 53 Summary For a succinct summary of the previous section, complete with definitions and the examples discussed, please refer to Table 1. 3.3 Role of Terms in Framework Thinking The terms in the framework and their definitions were chosen in a way to provide a different perspective on the standard PD process with which we are familiar. To help understand the role of these terms in the context of articulating the process, please refer to Table 2. While certain notions in this framework are fundamental (such as the role of design parameters and specifications), others such as the relationship between the architectural and specification domain and the role of constraints and feasibility in moving the process forward are non-trivial and will be discussed in further detail in the sections to follow. Table 2: Interpreting the Terms of the Framework Terms Role in Framework Architecture and Resolution Ways to think about ideas during the process Design Parameter and How we measure ideas (form and performance, respectively) Specification Architectural and The two solution spaces in which we work Specification Domain Instance, Point, and Set How we describe ideas in the aforementioned domains Constraint and Feasibility The mechanism by which the process moves forward (e.g. ideas narrowed, architectures embodied/detailed, etc.) 3.4 Use in the Literature This section reviews common uses in the engineering design literature of the terms outlined in Section 3.2. Since many of the terms presented here have been used extensively prior to this 54 work, I wish to review a non-exhaustive sampling of these prior uses. I also take the opportunity to point out if and how my definitions differ from established use, in addition to some of the implications of making such choices for the framework. Architecture Architecture is a notion that is central to the study of product development, and yet it is often informally defined in the engineering design literature. We shall begin this brief review by turning to two of the most influential educational texts on design theory: the works of Pahl and Beitz (Pahl et al. 2007) and Ulrich and Eppinger (Ulrich and Eppinger 2012), respectively. As these texts serve as many students' introduction to the product development process (thus impacting the perceptions they develop about the process early on), they seem to be an appropriate place to establish this review. By way of discussing the modularization of product designs, Pahl and Beitz define product architecture as "a scheme showing the relationship between the function structure of a product and its physical configuration." They go on to note that, "A product architecture can be used to describe the modularity of a product, which can be classified according [to] the functional and physical independence of its components" (Pahl et al. 2007). Ulrich and Eppinger take a similar view, albeit in different words, defining product architecture as "the assignment of the functional elements of a product to the physical building blocks of the product." They elaborate further, noting, "The purpose of the product architecture is to define the basic physical building blocks of the product in terms of what they do and what their interfaces are with the rest of the device" (Ulrich and Eppinger 2012). It should be noted that the study of product architecture was of particular interest to Ulrich, whose theoretical foundations on the topic helped him assert the importance of analyzing architecture in the context of understanding manufacturer success (Ulrich 1994). When comparing these two definitions, it is interesting to observe how both try to codify the relationship between function and form while defining "architecture." This is hardly unusual, where Jiao, Simpson, and Siddique provide a nice review of the literature on product architecture, overwhelmingly citing definitions whose primary focus is to relate function to form (Jiao, 55 Simpson, and Siddique 2007). Building upon my wish to present a more global sense of "architecture" to the PD research community, my definition of architecture decouples notions of function from the fundamental definition of architecture. Instead, the relationship with function is expressed via the relationship between the architectural and specification domains. The reasons for this choice will become clear in the chapters to follow. Specification Similar to how we reviewed use of the word "architecture," we look to prominent design methodology texts to gauge references to "specification" in the literature. Ulrich and Eppinger state, "We intend the term product specifications to mean the precise description of what the product has to do.. .A specification (singular) consists of a metric and a value" (Ulrich and Eppinger 2012). Unlike architecture, however, other leading authors have slightly different perspectives of "specification." Pahl and Beitz refer to a design specification as the collective list of requirements developed for a product design (Pahl et al. 2007). That is, while Ulrich and Eppinger's "specification" is meant to convert the list of customer requirements into actionable parameters for engineers, it is qualitatively different from Pahl and Beitz's "specification," which refers to the list of requirements in an abstract, qualitative sense. Among the aforementioned definitions, this work adopts a definition for specification that is closer to the Ulrich and Eppinger notion of the term. By treating a specification as a way to objectively measure a design's performance, we can liken reasoning on the specification domain to an optimization problem. This comparison will emerge in the following chapters. However, as first addressed in Footnote 7, Ulrich and Eppinger's sense of a specification also includes understanding of a target value for that specification, which my definition does not. To clarify, my definition reflects a variable which simply measures a performance parameter and does not address what value that parameter should take. Domains Domains play an important role in the design literature as they help compartmentalize the various streams of information that designers regularly work with during the PD process. Given the diverse array of pertinent factors to consider within the complex socioeconomic technical 56 systems in which we create and use the products we design, defining domains helps us to keep our thinking clear and organized. We begin by reviewing Pahl and Beitz's function structures, which are visualizations of sorts of the functional domain. A function structure helps translate requirements into an articulation of function, which by Pahl and Beitz's definition is solution-neutral (Pahl et al. 2007). The function structure will map inputs and outputs of a system, thus visually articulating constraints that a proposed form must satisfy in order to move forward in the process. In other words, in this visualization of the functional domain, constraints are plotted using the rules of generating a function structure, which then the designer can refer to when ideating concepts for form. The end result is that function informs form. This is consistent with the motivating idea behind Pahl and Beitz's framework, which is to distill the process of considering customer requirements and to conduct an ideation process that, as alluded to earlier, is solution independent. Suh arguably has different reasons to motivate his framework, better known as Axiomatic Design (Suh 1995). Suh's framework consists of four domains: the customer domain, the functional domain, the physical domain, and the process domain. These are represented in Figure 5. Suh takes us through this figure describing, "The domain on the left relative to the domain on the right represents 'what we want to achieve,' whereas the domain on the right represents the design solution of 'how we propose to satisfy the requirements specified in the left domain.' To go from 'what' to 'how' requires mapping." While a key tenet of Axiomatic Design is the "zigzagging" between domains to decompose a problem, it should be noted that Suh clearly presents a mapping of the functional domain onto the physical domain, which enables the designer to go from "'what' to 'how"' in his framework. Suh's framework is driven by his axioms for design, which on the surface differs slightly in perspective from Pahl and Beitz's focus on maintaining solution independence." It To clarify, following Axiomatic Design as it is intended should yield a process that remains solution independent. Nominally, however, Suh's framework is set up to confirm his observations of what makes good design, which are articulated via his design axioms. 57 mapping (CAs) Mapping {FRs) mapping [DMs (PVs) Customer Functional Physical Process domain domain domain domain Figure 5: Suh's Four Domains. Reproduced from (Suh 1995). Despite the minor differences between the approach of Pahl and Beitz and Suh, respectively, they both maintain a view of the process in which function informs form. This is certainly a valid perspective, but my contention is that this assumption does not adequately describe how designers actually behave during the front end of the process. Given our desire to generate a framework which honors the heuristic of considering more options, it is important to accurately describe the relationship between function and form in early-stage design in order to guide deeper insights into strategies that generate better, more novel outcomes. My perspective on this relationship is discussed in detail in Chapter 5. For more on the relationship between the functional and physical domains as it is discussed in the literature, please refer to (Schmidt 2015). Point, Instance, Set The terms point and set bear a legacy from Ward's original thesis describing his framework for Set-Based Design (A. C. Ward 1989). Given the context of his development of a mechanical design compiler that would automate a system's design from a catalogue of components, it was necessary for his use of the set, range, and related words to bear strict mathematical definitions in order for his design compiler to function. My purposes differ from Ward's original intentions in that I currently wish to describe, and not necessarily quantitatively model, a design process. Thus, 58 the terms point, instance, and set are defined in a qualitative manner, in contrast to Ward's initial approach. Before moving on, the reader should recognize that the term "instance" is not officially coined in Ward's writings. Instance is a term which I formalize in order to easily describe individual points that are generated for comparison throughout the process while maintaining the semantical contrast between the narrowing the initial "set" to the "point," or the final production version of a product. Constraints and Feasibility The concept of constraints is fundamental to the PD literature and transcends the relevant disciplinary bounds, bearing meaning in the greater engineering, economics, and related literatures. For such a versatile concept, it is important to trace its fundamental roots, and so we turn to a definition from mathematics. From Wikipedia: "In mathematics, a constraint is a condition of an optimization problem that the solution must satisfy. There are several types of constraints-primarily equality constraints, inequality constraints, and integer constraints. The set of candidate solutions that satisfy all constraints is called the feasible set" ("Surjective Function" 2015). As referenced earlier, conceptualizing the PD process as an optimization problem is a natural analogy and is the basis of significant literature across academic communities. From an optimization perspective, constraints define conditions that the solution space must satisfy, which is consistent with the definition for constraints presented in this work. It is interesting to observe how Wikipedia's definition for constraints ties in the concept of a "feasible set." Just like in the framework presented in this work, the idea of feasibility is hardly mutually exclusive from that of constraints-rather they are interdependent. Similar to Ward's optimization-inspired perspective, the role of constraints (and its interplay with feasibility) in the framework presented in this work is to gradually narrow the vast set of options down to the selection of a single point. The thoughtful propagation of these constraints is a central underlying theme to this work, and a proposed structure for approaching this problem is presented in Chapter 8. 59 3.5 Conclusion In this chapter, we introduce the lexicon for a new emergent framework for PD. We also review prior use of these terms in the literature, which enables us to draw distinctions from the framework presented in this work with other models conceptualizing the design process. For instance, our framework defines architecture in a way that is not directly tied with function, unlike many leading definitions for architecture in the literature. Furthermore, the domains as presented in our framework do not unconditionally accept the relationship between the functional and physical as it is often presented currently in design theory frameworks. In order to explore these differences further, we must first put the framework to use in describing the PD process, which is the focus of the next chapter. 60 4. Describing the Process Using the Framework In this chapter, I will use the terminology of the framework to guide descriptions of typical PD processes. This exercise will reveal deeper insights regarding the underlying mechanics of the process, leading to dedicated discussion on these points in the following sections. First, however, we shall begin with a generic description of the design process according to the framework. 4.1 Temporal Description of the Product Development Process Here, a series of assertions is made regarding the design process. These are made in bullet form in order to simplify the ensuing discussion and aid comprehension of the individual points. Clarification of certain points is made via sub-bullets. - The designer begins with a solution space of countably infinite potential architectures. o This assumes unlimited access to resources. I.e. the only active constraints at this point are laws of nature, preventing the sampling of certain concepts which violate these laws (e.g. an architecture that teleports). These types of architectures cannot be sampled, even with access to infinite resources. o The analogy for "countably infinite" comes from the probability of choosing a rational number between 0 and 1 along the number line (Nykamp). At this level of resolution, potential end outcomes are innumerable, though the space is still bounded by laws of nature, etc. " The process begins with the articulation of an architecture of arbitrary resolution. This architecture can be instantiated as groups of architectures at subsequent, finer levels of resolution. - The design process yields a series of instantiations in the specification domain from coarse resolution to fine resolution. " While architectures are natively described in the architectural domain, one can express an architecture of any resolution in the specification domain. 61 - According to this process, a specific instance at a coarse level may be understood as an architecture that embodies several instances at a finer level of resolution. " In the specification domain, the set of possibilities eventually narrows to the selection of a single point. This is the production version of the product. To help articulate the bulleted assertions above, please refer to Figure 6. represents an instance, each circle an Coarse Resolution Each star Fine Resolution architecture, and the collection of stars and circles within the parallel dotted lines represents the set. The parallel dotted lines themselves represent dormant constraints that are made active and are applied to the set, transforming it from coarser to finer levels of resolution. The first circle at the left of the diagram represents articulation of an architecture of arbitrary resolution (e.g. the concept of a V6 engine). This architecture four is instantiated by ++ + instances as shown in the diagram (e.g. four competing V6 engine concepts). Note n c +e that, consistent with its definition, an architecture can be expressed via one or more instances. As the process continues, Instance A rchitecture some architectures are eliminated from consideration while others are developed further via the propagation of dormant Set Figure 6: VisL alizing the Framework constraints that are gradually made active. 62 4.2 Other Visual Representations of the Framework In this section, I present alternative ways to visualize the framework in use, starting with one of the more prominent diagrams presented in American (and many international) engineering design education courses, Ulrich and Eppinger's generic PD process. An adapted form of this diagram (Ulrich and Eppinger 2012) is shown in Figure 7, upon which the terms of the framework are accordingly overlaid. Note that Ulrich and Eppinger's process begins with a "Planning" stage (the first converging funnel shown) followed by "Concept Development," which features subsequent divergence and convergence of ideas. While Ulrich and Eppinger do not explicitly label the pictorial portion of the diagram in their text, it is understood that the divergence and convergence depicted by the funneling out and in, respectively, represents the variance of ideas explored.' 2 Using the terms of the framework, this variance of ideas represents the set that is explored, as depicted in Figure 7. Similar to Figure 6, we can also label the embodiment of ideas from coarse to fine resolution. Finally, the set narrows down by the end of the process (the "Production Ramp-Up" phase in Ulrich and Eppinger's process) to the selection of a point, which is the production version of the product. While Figure 6 depicts the development of concepts and ideas in an abstract sense, perhaps it would also help to visualize this development in the specification domain, as is done in Figure 8. This visualization is carried out through three common intermediate phases of the PD process, employing terms used by Ulrich and Eppinger and Pahl and Beitz (some of the foremost authorities on presenting the PD process) (Ulrich and Eppinger 2012; Pahl and Beitz 1988). 12 This is implied from the chapters and discussion in the text on the phases of the process in question (i.e. Concept Development). 63 Point Coarse Resolution Fine Resolution Figure 7: Ulrich and Eppinger PD Process Interpreted Using the Framework. Adapted from (Ulrich and Eppinger 2012) As alluded to earlier, since the specification domain is where we measure project success, it is helpful to think about it as the domain which hosts design optimization. In Figure 8, imagine we are designing a product that only has two specifications of interest, thus yielding a twodimensional specification domain. Suppose the objective is to maximize the specifications represented on the horizontal and vertical axes, respectively. During concept development, architectures are proposed for consideration whose specification values span the blue amorphous space shown. However, given that the concepts proposed are qualitative in nature and lack deterministic definition of specification values, we cannot know the exact bounds of feasibility of the space of architectures explored at this point in the process. For this reason, the space shown is bounded by a black dotted line. During the system level/embodiment design phase, we sample three architectures which entail a subset of the original set explored during concept development. Since architectures are now going through embodiment, it is easier for us to quantify the range of specification values they entail, which is why the three architectures are enclosed by solid black lines. Finally, in the detail design phase, we select a group of instances entailed in the small red region within one of the three architectures explored in the embodiment phase. Within this red region, a single point, or the production version of the product, will be chosen. Note that the selection of a point from the red region during the detail design will not represent the theoretical optimum within the original set explored during the concept development phase (the optimum 64 would be the upper-right most corner of this original set, representing the maximum x and y values attainable). In practice, the designs we produce by the end of the PD process are often satisficed solutions (Simon 1996). Our inability as designers to understand the true bounds of feasibility during early stage design is a major factor behind such satisficed outcomes-an idea which will be explored in further detail in the sections and chapters to come. Concept Development Coarse Resolution Detail Design System Level/ Embodiment Design ) Fine Resolution Figure 8: Visualizing the Framework in the Specification Domain 4.3 Observations from Using the Framework Given that we have ended up with a group of instances that will lead to a satisficed solution in Figure 8, a natural question to ask is: what can we do to generate a more optimal solution? Fundamental to answering this question is an understanding of how we move from one phase to the next during the PD process. In other words, if we understood the mechanics underlying our transition from, for example, the concept development phase to the system level/embodiment phase, perhaps we could better understand why we end up proceeding with architectures that end up to be suboptimal solutions. Based off of the definitions of the framework and their use in our discussion and analysis of generic PD processes, we assert that the PD process progresses in the specification domain via the propagation of constraints. Regarding exactly how constraints propagate, we make the following observation using the descriptive lexicon of the framework: 65 "Most constraints in the specification domain are dormant pending the exploration of specific instances of architecture." In other words, in order for constraint propagation to begin, designers must begin to articulate form of arbitrary resolution. This serves to activate constraints which will begin to narrow the set of possibilities. Increasing resolution on these architectures will initiate further activation and propagation of constraints, eventually leading to the selection of a point-the production version of the product design. 4.4 Conclusion To summarize some of the critical observations over the past few sections, we have established that 1) the specification domain is where we measure product success, 2) the PD process is temporally driven by the propagation of constraints in the specification domain, and that 3) most' 3 constraints are not activated until the articulation of form. Given that form exists in the architectural domain and the propagation of constraints determining product success occurs in the specification domain, perhaps it would help to formalize the relationship between the architectural and specification domain. This is our task in the next chapter. 13 The most notable exception, as mentioned earlier, are laws of nature, which are active from the very beginning of the process. 66 5. Surjective Relationship In this chapter, I seek to explore the relationship between the two domains in this framework for PD. We begin with the articulation of the Surjective Theorem for Product Development, followed by discussion on and analogies regarding ways to think about the two domains. Finally, I conclude this chapter with additional insights from the surjective relationship that we can apply to other areas of inquiry in the design theory and methodology literature. 5.1 The Surjective Theorem for Product Development The Surjective Theorem for Product Development, hereafter referred to more simply as the "Surjective Theorem," is presented as follows: Theorem: "The relationship between the architectural domain and the feasible " specification domain is represented by a surective function.14 Here, it may help for the reader to have background on the definition of a surjective function. From Wikipedia: "a function f from a set X to a set Y is surjective (or onto), or a surjection, if every element y in Y has a corresponding element x in X such that f(x) = V' ("Surjective Function" 2015). A nice pictorial representation of a surjection is provided in Figure 9, where the sets A and B correspond to the sets X and Y from the Wikipedia definition, respectively. The black and red dots represent individual elements of sets A and B, respectively, and frepresents the surjective function. 14 It is important to note here that we are using the strict mathematical definition for function, and not one from the design theory and methodology literature. 67 S(=-A domain(A) f-+ eB range(A) Figure 9: A Surjection. From (Weisstein 2015). The Surjective Theorem tells us that for every feasible instance in the specification domain, there exists at least one corresponding architecture. Given the definition of feasibility, an instance in the specification domain is feasible if and only if an architecture exists to attain that instance. Therefore, the Surjective Theorem is a tautology, and so we take it to be true. To state it succinctly using the language of mathematics, we take the relationship of the architectural domain "onto" the specification domain to be true. Reasoning with the Surjective Theorem bears a number of interesting implications for PD and design theory, which we will discuss in the sections to follow. Here, I shall present perhaps the most important result from interpreting the theorem within the broader context of the framework presented thus far: " The specification domain is an artificial construct created by designers in order to organize their reasoning about product function, measuring product performance via individual specifications. Given its artificial nature, we can only explore the specification domain by mapping architectures directly onto it." This understanding can be obtained from a careful reading of the theorem within the greater context of the framework. One way to interpret this result is that since feasibility of (and thus existence of) instances in the specification domain are predicate on the existence of instances in the architectural domain, one cannot explore the specification domain without the articulation of architecture. 68 5.2 Understanding the Domains The purpose of this section is to help the reader understand how various aspects of the PD process can be categorized according to the two domains, helping to illustrate their interplay within the context of the surjective relationship. Please refer to Figure 10 for the ensuing discussion. Below the column headings of "Architectural Domain" and "Specification Domain," respectively, are analogies which we can use to interpret the role of the two domains. For instance, the architectural domain can often be thought of as the "design space" in which designers can directly modify variables that are pertinent to form, i.e. design parameters. The choice of these values in the design space, in turn, influence the outputs in the "performance space," i.e. the values of specifications in the specification domain. Accordingly, these labels correspond with notions of the architectural domain being the "physical domain" where form is clearly articulated, whereas the specification domain is the "functional domain" where product function is expressed and measured. It is important to note here that we are using the design theory and methodology sense of the term "function" (not the mathematical sense, which is employed in the Surjective Theorem). ArchitecturalDomain Surjective Function SpecificationDomain Design Space Physical Intrinsic Properties Projection Mapping Onto Performance Space Functional (DTM term) Extrinsic Properties e.g. Product dimensions Material choice Laws of physics e.g. Product weight Product cost Consumer demand Figure 10: Understanding the Two Domains Additional analogies for the architectural and specification domains include spaces for intrinsic and extrinsic properties, respectively. A simple example illustrating this distinction is 69 how a product's dimensions are inherently intrinsic properties of its form, classifying them as design parameters. Furthermore, dimensions themselves do not usually characterize performance. 15 A product's weight, however, is not necessarily intrinsic to the form itself, as it depends on the gravitational forces acting upon its mass (which is a function of its density and critically, its dimensions). While typical use cases for most consumer products do not necessitate the effects of variable gravitational forces, for designs such as space vehicles, these are extremely pertinent factors to consider. Moreover, across scales, product weight is often intimately connected to its performance, from handheld consumer devices to commercial aircraft. Another pair of examples include that of material choice and product cost. While most would agree that a design's material is an inherent property of its form, its cost is assigned from external markets for the material in question. In addition, cost is a critical specification for products, influencing among other things the consumer's decision to purchase the product. Of course, the examples above allude to the relationship between the architectural and specification domains, respectively, which is noted in Figure 10 by the arrow for "Surjective Function." Phrases such as "the architectural domain projected onto the specification domain" or "the architectural domain mapped onto the specification domain" all capture this relationship. To extend this language to the examples cited above, one could point out how a given product's dimensions can help map to its weight or how material choice can project onto a product's cost. In practice, specifications such as a design's weight or cost are outputs of a plethora of design parameters beyond dimensions and material, respectively. Thus, the role of the surjective function is to convert the inputs of design parameters in the architectural domain via a multiobjective function (e.g. factoring in influences such as gravitational forces and markets for the examples of product weight and cost) into the outputs of specifications in the specification domain. Using the 15 An example where individual dimensions may contribute to performance is a specification describing a product's footprint, with for instance the specific objective to minimize it. This is a fine distinction, as dimension intrinsically does not characterize product performance until a functional requirement such as the need to specifically minimize the size of the product is introduced. In this case, individual dimensions can act as specifications, as well. 70 aforementioned analogue terms, the surjective function makes the physical functional and the intrinsic extrinsic. While passive interpretation of the discussion above may suggest that design parameters are mutually exclusive from specifications, this need not be true. In some cases, the surjective function will take an input design parameter and directly output it as a specification without the influence of any other design parameters or outside factors. An example here may include product dimensions, which is discussed in Footnote 15. Finally, the last pair of examples listed in Figure 10 alludes to pertinent elements of the PD process that do not strictly qualify as variables in the respective domains. Rather, these elements still exert influence on the domains in the form of constraints. As alluded to earlier, the first group of active constraints to act upon the architectural domain are laws of physics. While these natural laws do not inherently characterize form (which is the role of design parameters), they help influence the values a given form takes, qualifying them as constraints. Furthermore, other factors can directly influence a design's specifications, classifying them as constraints in the specification domain. A typical example here is consumer demand. Preferences regarding desired cost, thresholds for minimum acceptable performance, and a host of other factors inform target values for specifications, limiting the set, or the space of solutions, to consider. 5.3 Additional Insights from the Surjective Relationship As we note earlier, the Surjective Theorem implies that for any feasible instance in the specification domain, there must exist at least one architecture that maps to it. Of course, this includes the possibility of many architectures that map to the same specification point. In other words, it is possible that distinct combinations of design parameter values (i.e. different instances of architecture) may yield the same instance or point in the specification domain. For some readers (particularlythose familiar with the optimization literature), this assertion may seem intuitively clear-they may skip the following paragraph if they wish. For others (including those who may not be convinced), a simple synthetic example illustrating this principle using the terms of the framework is discussed below. 71 Consider designing a two material alloy, where the only specification of interest is the cost of the alloy produced, and the design parameters of interest are the respective amounts of materials A and B to use. Suppose one is given a budget constraint of $16, and that the cost of material A is $4/lb. and the cost of material B is $1/lb. Without any other information, one could generate a trade-off relationship to represent the linear combinations of the weights of material A and material B, respectively, that would satisfy the budget constraint. Suppose the chosen weights of alloys A + and B are represented by the variables a and b, respectively. The relationship then, would be 4a b 16. Given that cost is the only specification of interest in this problem, if the designer wanted + to generate all architectures that would yield an alloy that cost $10, the expression would be: 4a b = 10, which represents all linear combinations of the weights of A and B, respectively, that attain that given specification value. Thus, from this problem, it is clear how distinct combinations of design parameters (i.e. different instances of architecture) may yield the same instance in the specification domain. 16 Returning to our original discussion, should one accept the possibility of one instance in the specification domain having more than one instance of architecture mapping to it, we can take this a step further. That is, should one be able to find a best or optimal instance in the specification domain, it implies the existence of at least one best architecture (defined by virtue of mapping to the best instance in the specification domain). This is a nice result, as it demonstrates how the Surjective Theorem supports conceptualizing the process as an optimization problem. Thus, while the methods underlying this work would not be ostensibly classified as design optimization, this 16 If, however, another constraint were to be added this problem-for instance defining the desired final weight of the alloy-the system of equations would yield one instance of architecture (which notably would still comply with the Surjective Theorem). For example, given the budget constraint 4a + b = 10 and a desired final weight of 4 lb (yielding the expression a + b = 4), then there would only be one architecture that satisfies this problem: an alloy of 2 lb. of material A and 2 lb. of material B. Should yet another constraint be added, we would illustrate the principle of linear programming in which the number of constraints cannot exceed the number of variables (or design parameters). This problem would be over-constrained and would not yield a solution or a feasible architecture. 72 result establishes how problems in the design optimization literature are compatible with and can be interpreted using the lexicon and structure of the framework presented in this work. Given our ability to contextualize design optimization problems, we can also use the Surjective Theorem to look the concept of Pareto efficiency in design in a new way. Suppose one had the ability to express the surjective function analytically (see Chapter 6). One could then contend that the Pareto frontier is in fact a subset of the codomain of the analytic surjective function. This assertion may be of particular interest to the engineering design community which has displayed an affinity for studying the Pareto frontier, and is addressed in further detail in the Appendix. Finally, another mathematical property of surjective functions is that the domain has a higher cardinality than the codomain. This has interesting implications for understanding the way designers generate ideas. Since these ideas are still under development, discussion on this point can be found in the Appendix. 5.4 Conclusion In summary, I introduce the Surjective Theorem, which relates the architectural domain and the functional domain via a surjective function. Furthermore, recognizing that the specification domain is an artificial construct, I assert that it can be explored only by mapping architectures directly onto it. We then explore analogies by which to interpret the two domains, their relationship, and the role that this domain mapping plays in the PD process. Finally, we close with using the results of our reasoning on surjection in the framework to contextualize other streams of work in the design literature, such as the study of design optimization and the Pareto frontier. 73 6. The Framework: Ideal vs. Practice Based on the framework presented thus far, in this chapter we shall present a vision for an idealized model for PD that takes advantage of the features and strengths of the framework. This idealized model is presented as a device by which to point out how the framework (and thus the PD process) differs in practice. These contrasts between the idealized model and the model in practice shall set us up for one of the central results in this work, discussed in Chapter 7. 6.1 An Idealized Model for Product Development To represent our idealized model for PD, imagine a computer program that accepts a design problem as an input and produces a production-version design (i.e. a point solution) as an output. One may think of this program as being an omnipotent, omniscient descendent of Ward's original mechanical design compiler (A. C. Ward 1989), which has been discussed earlier (see Chapter 2). In this idealized model, information would be complete (not asymmetric), static (not subject to change), and perfect (free from error). With the help of this program, we would be able to simultaneously interpret the plethora of variables that are parts of the greater socioeconomic technical systems in which our products reside. In this model, variables that are outside of the designer's direct control, such as material prices or consumer demand, are either fixed or have values that can be projected with 100% confidence to the time when the product is ready to use. A somewhat weaker sense of such certainty (though still effective and quite idealized) would be if the program were able to deterministically assess the quality of information it was handling. In other words, it could assign deterministic confidence bounds with full certainty of confidence level for the information being used in the model. 17 Armed with a self-awareness of the quality of the information it is processing, 17 This would make the model compatible with notions of fuzzy sets, which propose that the feasible bounds of the set cannot be assigned deterministically. Thus, while ranges of values which define the bounds of the fuzzy set cannot be known with full certainty, an idealized model would be able to self-assess confidence bounds on these boundary values with full certainty. 74 the program would always make the right design decisions at any given moment during the PD process with the information available to it. With variable uncertainty accounted for, the program would also benefit from knowing the analytic surjective function (i.e. the formal mathematical expression of the function between the domains). This would allow the program to perfectly map the architectural domain onto the specification domain, enabling optimization on the feasible set in the specification domain. Once the optimal point in the specification domain is found, with knowledge of the analytic surjection, this point could be easily mapped back to find the best candidate architecture(s). 6.2 Product Development in Practice, Described by the Framework There are many reasons why PD in practice differs from the idealized model presented in the previous section. Here, we shall focus on 3 of the major underlying reasons for such deviations, which are: 1) The analytic surjective function is unknown. 2) Uncertainty of variables and their effects on constraint propagation remain unknown. 3) A lack of understanding of the bounds of feasibility in the specification domain in early stage design. I shall discuss each of these reasons in detail in the subsections to follow. Surjective Function is Unknown In practice, it is very difficult to know the analytical function which maps architecture to specification. As noted earlier, the reason why this is useful or desirable is that if we can characterize an explicit mapping between the architectural and specification domains, we can optimize in the specification domain and backsolve (that is, find the architecture(s) that correspond to the desired instance or instances of specification). In fact, problems in the optimization literature operate on this assumption, as the objective functions of interest in such work can be thought to map subsets of the architectural domain to the specification domain (Yi, Shin, and Park 2008). 75 However, the question is: when along the process can we formalize this mapping, and is the assumed analytic function representing the mapping accurate anduseful? We can also address this question from the perspective of finding the Pareto frontier. Given that we have explored the potential for the Pareto frontier to be a subset of the range of the analytic surjective function, if we do not have knowledge of this formal surjection, when then can we describe the Pareto frontier? In practice, we can obtain a representation for the frontier via trial and error. However, this method is limited in that bounds of feasibility often must be established first (discussed in further detail below). Furthermore, given that trial and error is involved, there will be a certain level of fuzziness or uncertainty when describing the Pareto frontier. Many optimization algorithms use a heuristic to define convergence criteria which defines when to stop searching for the frontier. Thus, the proposed frontier may not truly match with the true optimal idealized frontier that is a subset of the range of the analytic surjective function. This may be deemed acceptable for many applications, but it is not the true optimal subset of solutions that is promised by the Pareto frontier. Another way of interpreting the issue that the analytical surjective function is unknown is via the lens that designers cannot directly engage the specification domain without first projecting architecture (an assertion made in Chapter 5). Once even low-resolution architectures (e.g. concepts which may not have been embodied yet) are articulated and then mapped onto the specification domain, then trade spaces can be explored and optimality of specifications can be pursued (Kerga 2013). For instance, consider designing an I-beam for a structure. Given our knowledge of mechanics, materials, and other areas, we can analytically optimize many performance specifications of the beam without having to develop an I-beam in real life. However, before we can exploit analytical relationships from the applied sciences to address our problem, we must first decide that it is an I-beam that we want to design-and thus we must formally articulate this concept of architecture and map it to the specification domain before exploring optimal specification values. The question then becomes: what is done at the earliest stages of the PD process, before any concepts of architecture have been articulated? How can we explore the specification domain at this point? The simple answer is that we cannot. In the absence of knowledge as to how the 76 two domains map to each other, in early-stage design, the prescribed solution in practice is to prototype rapidly and cheaply. As noted earlier during the review of the characteristics of setbased thinking (see Chapter 2), authors have noted how this practice of early, rapid, cheap prototyping is done to promote learning of the system and the design problem at hand. To put it in the terms of the framework, prototyping is done to understand the behavior of the surjection in the absence of any formal knowledge of its expression. Uncertainty of Variables In addition to a lack of knowledge of the surjective function (and its behavior until form is expressed and mapped to specification), in practice we also lack perfect information on the pertinent variables in our system. Changes in variables such as customer requirements or market demand, regulations, the intellectual property environment (e.g. new patents), supply chains, and other factors present a significant challenge for designers. Often, these changes can compound across systems, where a seemingly small change in one parameter at the subsystem level may propagate to other subsystems, causing notable consequences at the system level. Such change propagation can occur regardless of the scale of the system, though the effects can be particularly profound in large, complex systems. Results of these changes include rework, excessive iteration, cost and schedule overruns, and other deleterious effects (Helms et al. 2014). Other reasons for uncertainty in variable values include the low resolution of ideas early on in the process and issues associated with information quality. As discussed earlier, ideas progress from coarse to fine resolution throughout the process, and until many details across the product one is designing are worked out (such as defining subsystems and their interfaces (Ulrich and Eppinger 2012)), one cannot progress to detail design to lock down the ranges of design parameters and corresponding ranges of specifications to mitigate such uncertainty. Regarding information quality, designers in practice often process, unbeknownst to them, imperfect information that is laden with error. Designer mistakes committed due to such faulty information can lead to grave consequences, such as the Mars Climate Orbiter incident (Oberg 1999). As discussed in Chapter 2, a key tenet of SBD is to leave options open as long as possible in order to mitigate or avoid the aforementioned consequences of rework that can result from variable changes that occur during the PD process. In a sense, it is a paradoxical strategy as it 77 suggests that variable uncertainty can be dealt with by embracing that very uncertainty and to delay convergent decision making which would lock down certain parameters that would be sensitive to unexpected changes. While certain practitioners of SBD have claimed to have successfully mitigated the effects of variable uncertainty with the method, the fact remains that SBD's prescribed approach goes against conventional wisdom in the PD community, potentially influencing its lack of broader adoption as a methodology in industry. That is, in spite of any evidence that leaving options open for longer may be beneficial, most designers are not very good at internalizing all this uncertainty and would much rather lock down aspects of a design as soon as possible. In fact, when the cost of iteration is cheap (such as in software design), this is the preferred strategy in practice (see Section 2.3.1 for further discussion). For further discussion on suggested approaches to manage uncertainty of variables, see Chapter 8. Unknown Bounds of Feasibility Of course, a lack of understanding of the mapping between the architectural and specification domains and the uncertainty of variables (in both domains and from external factors) implies that we cannot predict the bounds of feasibility in early-stage design. The reason why this information is pertinent, as addressed earlier, is that the pursuit of optimality often occurs at the bounds of feasibility. While these limits may be predicted analytically, in practice such analysis is difficult to rely on due to the compounded uncertainties of domain mapping and variable uncertainty. Thus, the solution in practice to understanding the bounds of feasibility is simply to try things out. Historically, the development of entire industries that traditionally operate at the bounds of feasibility, such as aviation and pharmaceuticals, has progressed via relentless experimentation to understand what works and what does not (Thomke 2003). A Common Theme: Articulating Form As we have noted thus far, these issues are interdependent, and one can be described as being a consequence of the other and vice-versa. Thus, it should not be surprising to consider that the proposed solution in practice to each of these quandaries is simply to articulate form. Whether it is prototyping to understand the behavior of the surjective domain mapping, locking down design parameters to mitigate the uncertainty of variables during the process, or experimenting in order to understand the limits of feasibility, each of these strategies essentially amount to the designer 78 expressing form before engaging in any earnest exploration of the functional or performance space. While this may seem to be an innocuous observation at first blush, the significant implications of needing to first articulate form are discussed in detail in Chapter 7. 6.3: Conclusion In this chapter, we present an idealized model for PD using the framework, which is then used as a way to contrast how PD in practice differs. Discrepancies between this ideal vision and how PD is actually practiced can largely be attributed to: 1) the analytical surjective function being unknown, 2) the uncertainty of variables in the PD process, and 3) the bounds of feasibility being unknown. In practice, designers manage these challenges through a variety of activities which result in first articulating form-the consequences of which are the topic of the next chapter. 79 7. The Designer's Dilemma The following chapter ties together observations from the previous chapters to deliver one of the important observations of reasoning with the framework: The Designer's Dilemma. I present the formal articulation of The Designer's Dilemma in Section 7.1, followed by context for the "Dilemma" in the design literature in Section 7.2. In Section 7.3, I illustrate examples of the effects of the Dilemma in the widely popular sport of football. In Section 7.4 I delve into some insights we may obtain from reasoning with this result, followed by strategies to address the Dilemma and concluding thoughts in Sections 7.5 and 7.6, respectively. 7.1 The Designer's Dilemma Up until this point, we have used the framework to guide a variety of insights on the process, such as how product success is measured in the specification domain, yet it cannot be explored prior to a basic articulation of architecture. Furthermore, in the previous chapter, we summarize the numerous ways in which we as designers are unable to internalize inherent uncertainties of design in practice (see Section 6.2) and how the solution is to articulate form. Tying together the various prior reflections on the role of expressing form in the PD process, I am now ready to present one of the paradoxes of Set-Based Thinking and innovation in early-stage design in general, the Designer's Dilemma, which is expressed as the following: "One cannot reason on the solution space of potential designs without some articulation of form. However, by virtue of articulating form, due to bounded rationality, one is inherently prone to some fixation." The first sentence should follow from prior chapters, just that we have generalized the language to "reasoning on the solution space of potential designs," noting that potential designs are for practical purposes measured and compared in the specification domain. The second sentence alludes to a concept that was briefly mentioned during the review of the literature in Chapter 2-fixation. Depending on one's domain, the term "fixation" can support a variety of meanings and interpretations. The notion of fixation expressed in the Designer's Dilemma is closest to the definition, "a preoccupation with one subject, issue, etc.; obsession" ("Fixation" 80 2015). Within the context of design, we can interpret fixation as "being preoccupied or influenced by the nature ofprior-articulatedconcepts or ideas." In Section 7.2 we shall review definitions for fixation as they are cited in the design literature, which differ slightly from the notion we described in the prior sentence. Another concept in the Designer's Dilemma that has not yet been formally introduced in this work is bounded rationality. In The Sciences of the Artificial, Simon suggests bounded rationality as a way to understand how humans can only make decisions within the realm of their cognitive limits (Simon 1996). Given the context of bounded rationality, Simon argues that in practice, optimal solutions to design problems are hard (if not impossible) to realize, and that most solutions are in fact "satisficed." In describing this phenomenon, he states: "An earmark of all these situations where we satisfice for inability to optimize is that, although the set of available alternatives is "given" in a certain abstract sense.. .it is not "given" in the only sense that is practically relevant. We cannot within practicable computational limits generate all the admissible alternatives and compare their respective merits. Nor can we recognize the best alternative, even if we are fortunate enough to generate it early, until we have seen all of them. We satisfice by looking for alternatives in such a way that we can generally find an acceptable one after only moderate search." This short passage insightfully reflects on a number of items that are relevant to our framework. Translating this passage accordingly, Simon points out that while the set is given abstractly (i.e. we know what types of constraints are relevant and bounding the space), we do not know the practical limits of that set (see Section 6.2 for further detail). Furthermore, it is implied that to have any "practically relevant" knowledge, alternatives must be generated and comparedwhich is another way to express the articulation of form. Finally, Simon argues that we cannot "recognize the best alternative.. .until we have seen all of them." Thus, given the "practicable computational limits" of our minds, we must simply try out options, finding "an acceptable one after only moderate search." Given that we established that we need to generate options for comparison (i.e. articulate form) within the limits of our bounded rationality, my intent is to take this one step further, suggesting that in practice our cognitive limits are also responsible for generating fixation (see my 81 definition above) among those options or ideas generated for comparison. The implications of this assertion are broad-some of which shall be addressed in Section 7.4-but for now, it is important to appreciate the conundrum this presents for innovating in early-stage design, especially within the context of trying to consider more options. 7.2 Contextualizing the Designer's Dilemma in the Literature One of the earliest studies of fixation in the engineering design literature is from Jansson and Smith, who define fixation as "blind adherence to a set of ideas or concepts limiting the output of conceptual design" (Jansson and Smith 1991). Their studies are largely descriptive, studying the performance of engineering design students and professional design engineers in generating as many designs as possible in response to a given design problem. Priming the experimental or fixation group with an example design solution to the problem, they consistently found poorer output of ideas (in terms of quantity generated and uniqueness of ideas) from the fixation groups relative to the control groups that were not primed with the example solution to consider. In the years following Jansson and Smith's pioneering study, many authors in the design community have studied fixation, notably including Linsey et al.'s multi-institution collaboration on the topic (Linsey et al. 2010). In their study, Linsey and colleagues confirm the results of Jansson and Smith's studies among a group of academics who teach design (and ostensibly are aware of the phenomenon of fixation), in addition to suggesting defixation strategies (which shall be addressed in Section 7.4) and measuring the study participants' perceptions of the degree to which they fixated. Notably, in the aforementioned studies and others (a nice review is provided in (Linsey et al. 2010)), fixation was studied in a controlled experiment, where the experimental group is baited or primed with a solution or concept to the problem they are asked to solve, whereas the control group is simply given the problem without any bait for explicit fixation. The quantity and variance of ideas between the two groups are compared, thereby allowing researchers to assert the effects of fixation in the experimental group. Nevertheless, thus far in the literature, fixation is presented and studied in a relative sense. That is, by virtue of comparing the poorer performing experimental 82 group to the control group, one can assert the presence of fixation. My contention, however, is that the control group also suffers from fixation (albeit to a relatively lesser degree). Given that we have established the need to articulate architecture in order to explore the performance space and drive the process forward, and that bounded rationality limits our ability to generate and compare a large number of options, we fixate on the architectures that we explicitly generate for comparison. By virtue of articulating an architecture, we simultaneously 1) ground the abstract problem at hand by giving our minds a practical solution to consider and 2) increase the chance that our thinking on generating the next concept(s) will be influenced by the nature of the architecture that was just articulated. Thus, in lieu of Jansson and Smith's definition for fixation as "a blind adherence to a set of ideas or concepts," my argument is that fixation occurs without necessarily having a "blind adherence" involved. Despite Linsey et al.'s interest in studying the perceptions of fixation post-experiment, I assert that one can be fully aware of their fixation, and it does not ultimately matter. Fixation is a direct consequence of the way we reason about design within the limits of bounded rationality-and thus far in our cognitive development as a species, it is inevitable. This is one way to appreciate the full extent of the conundrum that is the Designer's Dilemma. 7.3 An Example from Football The Designer's Dilemma can be interpreted in a variety of domains outside of traditional PD. Perhaps one of the most compelling comprehensive examples by which to interpret the Dilemma coupled with the desire to consider more options comes from American football. I choose this example due to the depths of the analogies which can be drawn to the PD process, especially in context to how it is presented in this work using the framework. In football, between downs, the objective of the offense is to choose a play that will maximize yards gained in hopes of scoring points. Once in play, the opposing team's defense seeks to stop the advance of the offense in hopes of securing possession of the ball for their team. Prior to the snap, the offense chooses a specific play from its playbook to run after the snap. In other words, the selected play is a point from the feasible set-which is the team's playbook. Once the offense begins its play, the defense must have the appropriate formation of react to that play. 83 Constraining the initial set of possibilities for the offense (i.e. its playbook) includes the individual capabilities of members of the offense, which can be thought of as constraints on the form of plays articulated, thus acting as constraints in the architectural domain. Other factors include the strengths and capabilities of the defense and weather conditions during time of play (e.g. strong winds, slippery turf from rain, etc.). These factors influence the performance and success of the offense's chosen play, and thus they can be thought of as constraints in the specification domain. In hopes of maximizing success of the play, some teams take a Set-Based Thinking approach to play calling-the no-huddle offense. In a no-huddle scheme, the quarterback or playcaller (if not the quarterback, the offensive coordinator or head coach) often reads the defense based on its formation prior to and during the play and choose an appropriate play for maximum effect (Kelly 2013). That is, instead of committing to a point or play prior to the snap, the playcaller will consider more options by allowing the defense to constrain the set of options or plays available based on its formation in hopes of exploiting the defense's weaknesses during the play. Thus, the defense will initiate the process of articulating the final architecture of the offense's play by limiting the initial set of feasible plays from which play-caller may choose. In practice, the no-huddle offense is very difficult to execute and requires many Characteristics of Set-Based Thinking to enable. For instance, a no-huddle offense relies on a tolerance for under defined specifications, as the quarterback often will not commit to a specific point or end outcome to the play until well after the snap. That is, the quarterback will inform his team of the low-resolution play which he intends to run, which can have a variety of end outcomes depending on how events transpire. Reading the field while.the play is in progress, the quarterback will exploit specific matchups he sees between, for instance, a receiver and a defender, and then commit to a specific point or end outcome to that play. Other Characteristics required include more efficient communication among subsystems, subsystem exploration of optimality, and a support for flow-up knowledge creation. In the absence of the huddle, players (i.e. subsystems such as the receiving corps and offensive line) must efficiently communicate with one another at the line of scrimmage with code words and signals. Subsystem exploration of optimality and support for flow-up knowledge creation go hand-in-hand in this context, as the quarterback 84 (another subsystem in the context of the greater offense) has considerable autonomy in choosing the final play and changing the original call at times before and after the ball is snapped. Thus, we have established how the no-huddle offense, if executed properly, can take advantage of a situation by allowing the defense to initially constrain the solution space of plays in an effort to exploit the defense's weaknesses. However, part of the challenge here is that an offense that only reacts to the signals of a defense can be exploited. In other words, an offense that only chooses plays by reacting to the signals of the defense becomes fixated on a certain subset of plays designed to react to those signals. A good defense will exploit this fixation by periodically differentiating signals of the play to occur from the actual plays that are run. For instance, if the defense realizes that the offense will run only in reaction to a passprotection formation, it may give the signals of conducting the same pass-protection play when in fact it has plans to defend against the run. To overcome this, an offense must not become fixated on run-only plays in reaction to such signals, and must be willing to explore larger portions of the solution space beyond handing it off to the running back to rush (i.e. the typical "run" play). 7.4 Observations from Reasoning with the Designer's Dilemma We can use the Designer's Dilemma to make a number of interesting assertions about the PD process, which shall be detailed in the respective subsections to follow. Interpreting Concept Development as Indiscriminate Search Since we cannot directly reason on the specification domain without expressing architecture, and given the cognitive limits imposed by bounded rationality, concept development , (and any prior phase such as opportunity identification) essentially amounts to indiscriminate' 8 18 1 avoid using the word "random" due to its connotations from a probability perspective. A truly random search method would generate instances whose aggregated specification values could be understood as being sampled from a normal distribution, for instance. Fixation implies that prior articulated instances will influence the articulation of future instances, which in probabilistic terms would decidedly represent a non-random process. 85 inefficient search. Without knowledge of the analytical surjection linking the two domains early on in the process, much of the reason to articulate form is to test the viability of a concept in the specification domain. This is why prototyping is often thought to promote learning early in the process. The result is indiscriminate search of the specification domain where we are attempting to find architectures or concepts that are qualitatively better than the ones we have come up with so far. The search is clearly inefficient since we do not have any analytical tools at this point to directly assess optimality in the specification domain. As we note earlier, once there is sufficient resolution on the concepts expressed to create an approximation of the surjective function, we can apply analytical tools from our knowledge of optimization techniques to find appropriate solutions. However, once we reach this point in the process, the set we are exploring has already been subjected to the biases of fixation from earlier. Of course, this point segues nicely to our next observation. Why Novel Outcomes in SBD (or any other method) are a Chance Pursuit Earlier in this work, we identified a latent issue to come out of the SBD literature, which was the notion that in practice, SBD as a process does not necessarily guarantee novel outcomes. Now, after our careful reasoning with the framework presented in this work, this result should not be surprising. As a project management method, SBD's strength lies in requiring designers to explicitly justify why certain solutions are better than others. This is most effective in the form of trade studies, where the specifications of options have been quantified, generating "trade spaces" which are often methodically plotted and analyzed to identify the best alternatives (Oosterwal 2010; Kerga 2013). However, trade studies do not offer any answer for the fuzzy, unquantifiable front end of the process. A minimum level of resolution of architecture is required to conduct these trade studies, which implies the inadvertent elimination of a number of options in order to get to that point. Thus, it is a somewhat disingenuous notion for practitioners of SBD to quantify the number of options available in the initial set and indicate their reasoned elimination of this initial set to the eventual selection of a single point solution ("Ship-to-Shore Connector Implementation of Set-Based Design" 2009). Not only is this practically infeasible due to bounded rationality, but it does not describe what actually happens-that is, the theoretical number of 86 options in the initial set are never brought into the trade studies phase of the process when options can be quantified and directly compared. As we note earlier, once we are ready to approximate surjection (that is, we can quantify specification values on options), it is too late, since we are conducting trade studies on a subset of the original theoretically countably infinite set that has been subjected to our inadvertent fixations from earlier. To provide a simple example inspired from a class project completed during the course of this thesis, consider a company is designing a diesel generator in a particularly price-sensitive market. Suppose this company's generator is cheaper than most competitors, but it has received complaints from customers for being too noisy. The company, driven by its desire to retain its high market share as one of the cheapest options on the market, decides to conduct a variety of studies to determine how to minimize the noise of the generator. They experiment with different shaped enclosures (to minimize transmission of sound from the components inside), different sound absorbent materials to line the enclosures with, and different mounts upon which the engine sits (to minimize vibration and its influence on noise emitted). Any solutions that the company's engineers propose to management must be within the cost of the current generator. While the engineers occupy themselves with a variety of trade studies, engineers and management lose sight of the fact that instead of redesigning the components that they manufacture for the generator, perhaps they could focus on acquiring better parts that they do not design, such as the engine. As it turns out, the engine is responsible for producing the most noise in the assembly. It is also relatively underpowered compared to competitors. Thus, while being fixated on maintaining current price point, management had missed the opportunity to explore the option in investing in a better engine that was quieter and more powerful and passing on the cost to consumers. Management has implicitly assumed that consumer demand is highly elastic (i.e. sensitive to price hikes); however, with performance improvements such as noise reduction and increased performance from the engine (e.g. power output), perhaps the consumer would be willing to embrace the higher cost. Despite potentially losing the most price-sensitive consumers this way, if enough consumers are retained in spite of the price hike, then the strategy will be profitable for the company. 87 Taking a step back, however, perhaps the key to increasing profits for the company has nothing to do with focusing on noise, power output, or attaining a low price for current generator offerings. Perhaps the company is actually fighting poor brand perception among consumers in the target market. Nevertheless, without first articulating this as an option, management cannot dedicate resources to determine whether this is a viable option worth investing in (e.g. focusing on rebranding the company in this segment). As SBD would prescribe, no potential trade studies can be conducted without first deciding what strategies should be quantified and compared. What this means is that if designers are simultaneously astute and serendipitous with the articulation of the concepts they want to consider during concept development, then there is a chance to obtain a novel outcome at the end of the process. More often than not, though, designers will not strike this innovative gold. This is why despite following a methodical process such as SBD (or any other known PD technique, for that matter), novel outcomes can never be guaranteed. Furthermore, it is another way to understand why innovative products in practice are difficult to come by and why most PD projects end in failure. Why Artificial Intelligence Does Not (Yet) Pose a Threat to Human Designers A lack of understanding of the surjection between the architectural and functional domains in early stage design poses a challenge for the development of artificial intelligence. While a computer has the ability to trial-and-error and generate many more alternatives for comparison than a human, it does not have the ability to analytically out-reason humans. In the terms of the framework, like a human, a computer will not have the ability to surmise the surjection without the articulation of form. Unlike a human, however, a computer cannot articulate form without a problem that has been appropriately scoped for it. To successfully decompose an ill-defined problem and its uncertainties and to propose solutions to such a problem in the early stages of the PD process is a unique effect of human ingenuity whose underlying cognitive mechanisms we do not yet fully understand. Our inability to create an algorithm to emulate human reasoning in early-stage design has thus far limited the potential for artificial intelligence. Ward's original PhD thesis describing SBD was motivated 88 from an artificial intelligence perspective-to create a "design compiler" that would automate decision-making on assembling a system from a catalog of individual components (A. C. Ward 1989). Ward's pursuit was quite relevant at the time, where the development of artificial intelligence was seen as a key area for growth in the design theory and methodology community (Finger and Dixon 1989). In the intervening years since Ward's foundational work, it is interesting to note the community's decreased focus on developing artificial intelligence for the purpose of removing the human designer from many contexts. Algorithms developed today are largely intended to complement human decision-making as opposed to replacing it altogether. Without analytical tools to augment the level of artificial reasoning, despite being much quicker, computers will only remain as intelligent as the humans that create them.19 7.5 Strategies to Counteract the Designer's Dilemma Given the established inevitability of the Designer's Dilemma, what can be done to mitigate its effects? In this section, we sample existing and proposed techniques to reduce the effects of the Dilemma, which can be broadly grouped into two categories-those that amount to "blind search" and those that embody "informed search." Blind Search As we have established the indiscriminate and inefficient nature of the concept development phase, it should not be surprising that the vast majority of proposed ideation methods (including many documented defixation strategies) amount to blind, uninformed search. That is, the emphasis of these methods is to generate sheer volume of ideas, without particular attention to whether diverse regions of the solution space are explored. Many divergent concept generation techniques, such as the morphological matrix and the 6-3-5 technique essentially amount to indiscriminate search where quantity of ideas generated is the heuristic for successful ideation (rather than their ability to sample larger parts of the unexplored solution space). Some methods 19 For further discussion, there exists significant work on the potential for artificial intelligence to equal ("Artificial General Intelligence" 2015) or surpass ("Superintelligence" 2015) human intelligence, including criticisms of these viewpoints ("Chinese Room" 2015). 89 such as the 6-3-5 technique may in fact promote fixation by encouraging certain members of a team to modify another's idea, without any sense of whether that idea is superior to more diverse concepts that have yet to be proposed or explored (Pahl et al. 2007). A number of methods have also been proposed with the explicit goal to promote defixation. Linsey et al. review several of these techniques, which can be characterized as promoting "design by analogy" (Linsey et al. 2010). These include, for instance, considering analogies from the natural world and biomimetics, and database-driven tools which suggest analogies for a design based on designs from other domains that utilize similar functions or address similar customer needs. As Linsey and colleagues note, these methods may help designers frame their problem in a new way, potentially enabling more novel proposed concepts and solutions. While refraining a design problem has been shown to potentially lead to better proposed solutions, it does not fundamentally alter (and thus it does not improve) the mechanism by which form is articulated. The non-random process by which form is articulated is only marginally improved when presented with framing techniques that may encourage designers to sample new, unexplored subsets of the initial set of possibilities in the specification domain. Despite this, these reframing defixation techniques cannot guarantee that 1) further fixation will not occur on the new subspace/region, and 2) that new subspace is far enough from the subspace of the prior articulated idea to be worth sampling (otherwise the specification values or performance of concepts from both subsets will be very similar, defeating the purpose of the defixation exercise). Informed Search Given the aforementioned drawbacks of blind search, the question becomes: can we more deterministically ensure more diverse sampling of the specification domain to truly embody the ethos of considering more options? This is the goal of what I term to be "informed search." With the understanding that optimality often occurs at the bounds of feasibility, one method may be to intentionally articulate form in order to sample the limits of feasibility. A theoretical basis for such methods has been provided by Malak and colleagues in the context of finding the parameterized Pareto frontier (Malak and Paredis 2010). Motivated from a set-based perspective where subsystems may explore optimality, this work seeks to articulate knowledge of Pareto 90 frontiers in the absence of a well-defined higher system level objective function. From the search algorithms explored in this line of work, we could potentially create ideation techniques which act in a similar way by intelligently searching for the bounds of feasibility as opposed to blindly articulating new instances with the hope that enough have been articulated and compared to be confident in having found a subset of superior solutions to carry forward through the process. An additional property of a surjection is that the domain has the same or higher cardinality than the codomain (see Appendix for further discussion). Thus, another strategy for informed search may be to understand the sensitivity of variable mapping between the architectural and specification domains. That is, for a given design, changes in which design parameters lead to the most significant changes in specification values? In this sense, we can directly manipulate how we articulate form in order to attain broader sapling of the specification domain, as opposed to refraining techniques, where we are generally unaware as to whether the technique actually aided us in sampling new subsets of the specification domain until we try it out and measure it (such as via prototyping). Understanding inter-domain variable sensitivities remains' an area for potential future work. Of course, the critical reader may notice a critical caveat to these strategies for informed search. In order to create search algorithms inspired by the search for a parametrized Pareto frontier or exploiting inter-domain variable sensitives, one must have some initial working knowledge of the behavior of the surjection between the architectural and specification domains. Thus, at the earliest stages of the process, one is still compelled to begin the exploration of the specification domain via the blind articulation of form, just as the Designer's Dilemma tells us. While the "blind" and "informed" techniques addressed in this section certainly have the potential to promote some defixation, they cannot guarantee such success in early-stage design. 7.6 Conclusion In this chapter, we tie together the cumulative observations of previous chapters from reasoning with the framework to articulate one of the paradoxes of early stage design: the Designer's Dilemma. Contextualized from our understanding of the bounded rationality of human cognitive processes, it recognizes the inevitability of fixation as a result of being forced to 91 articulate form in order to begin exploration of the specification domain. A few results from reasoning from the Dilemma include interpreting concept development as indiscriminate, blind, inefficient search, and recognizing why Set-Based Thinking cannot guarantee novel outcomes. Finally, we explore a few strategies meant to counteract the effects of fixation while observing the limitations of these techniques. 92 8. Understanding Constraint Propagation In this chapter, I propose a different way to conceptualize and manage uncertainty in the PD process and minimize rework. In Section 8.1, I establish the argument for understanding an order for constraint propagation, and in Section 8.2, I present a framework for doing this that applies reasoning from the study of econometrics. It should be noted that this chapter is largely experimental and it is intended to propose groundwork for future study on this topic rather than providing answers on the topic of constraint propagation. 8.1 Identifying an Order for Constraint Propagation As we note earlier in this work, the propagation of constraints drives the process forward. Recall from Section 4.3, "Most constraints in the specification domain are dormant pending the exploration of specific instances of architecture." Via the "exploration of specific instances," dormant constraints are slowly made active. While we have defined "dormant" and "active" constraints earlier in this work (see Section 3.2), it is important to note that these are not binary conditions. Similar to how an architecture is developed from coarse to fine resolution, constraints undergo a similar process whereby they transition on a spectrum from "dormant" to "active." For example, constraints due to manufacturing process employed on a part may gradually activate by first identifying suppliers to use and their capabilities (which limits the set of potential processes that could be used), followed by choosing a particular manufacturing process for the part and setting the parameters for that process. Thus, the initially dormant constraints affiliated with manufacturing process employed will continue to emerge until they are fully active during the actual production of that particular part, helping increase resolution on the part's design from low (e.g. before suppliers are identified) to high (e.g. after the part has been manufactured). Interpreting constraint propagation as the mechanism which moves the process forward, perhaps there is a preferred order in which constraints should be gradually activated. Understanding this order may help avoid rework, change propagation, and other deleterious effects associated with the uncertainty of various parameters in PD in practice as discussed in Section 6.2. 93 A rudimentary example for a framework to order constraint propagation is provided by Day in his Real-Win-Worth It (RWW) framework (Day 2007). Surmising a preferred order for constraint propagation implies that there are certain parameters or variables that if modified would cause larger systemic effects than others. For instance, from an RWW point of view, constraints pertaining to "is the market real?" such as the size of the potential market and consumer demand at a certain price range take precedence over downstream manufacturing constraints such as whether existing supply chains can produce the proposed product. In other words, while both types of constraints (i.e. market and manufacturing) clearly influence the feasibility of whether a product can be produced, most would agree that for a profit-maximizing firm, it does not make sense to produce a product that is technically feasible to manufacture but for which there is not sufficient consumer demand. Thus, for most20 profitmaximizing producers, market constraints should be applied to the set before manufacturing constraints to avoid the outcome of producing a product for which there is no significant market and which will cause the firm to lose money. 8.2 An Econometrics Inspired Framework for Ordering Constraint Propagation Given the innumerable interdependencies within the socioeconomic technical systems in which the products we create reside, it may help to analyze this problem from a social science perspective. Thus, I propose the creation of a model for the product development process that is similar to an econometric model, with explanatory and response variables. Isolating explanatory variables in such a framework would help isolate which variables may contribute to larger propagated change throughout a given design. 20 This assumes the proposed product is not part of a larger strategy for the firm to for instance gain market share or build upon its reputation, in which case losing money on a per-unit basis for a particular product may be a profitable strategy in the long run for the firm. 94 In this framework, response variables would be those whose values depend on those of explanatory variables. Explanatory variables are those that "explain" the values of the response variables. For instance, an explanatory variable may be material choice for a part, whereas a corresponding response variable may be the cost of that part. However, these causal relationships cannot be established until the cumulative effect of all pertinent variables are accounted for. As prescribed by standard econometric theory, one may ascertain causality among variables by studying a variety of regressions among pertinent variables in the model. Given the plethora of variables involved (both in a traditional economic model and in a hypothetical model for PD), the simultaneous influences of many variables presents a non-trivial challenge for determining which relationships can be interpreted as causal as opposed to those which are simply correlated. At this point, the reader may be asking: how can such an analysis be carried out? One approach is to collect historical data on a portfolio of past PD projects that are all similar in nature (e.g. similar products/applications). For each project in the portfolio, quantitative data on pertinent variables are collected (e.g. material cost, product sale price, etc.) and then inserted into a model upon which multiple regressions may be conducted. From these regressions, explanatory variables among the historical data on the projects could be identified and compared across different cases within the portfolio of past projects. The comparison would yield insight into which variables are most often explanatory across these projects. The hypothesis is that these variables, if subjected to change in PD projects that are similar to those that were studied for the regressions, would cause the greatest perturbations or largest effects across a system by virtue of their explanatory nature. By observing which variables are most often explanatory, we can begin to map this knowledge back to a temporal understanding of which constraints should be activated before others in order to avoid change propagation, rework, and other issues that may be mitigated should constraints have been propagated in the proper order. There are still a number of issues to keep in mind prior to applying an econometric modeling-inspired approach to studying PD projects. For instance, certain variables can be shown to be explanatory, knowledge of which may not be of ultimate use to the designer. Considering the example of an engineering drawing (discussed in Section 3.2 in the context of the architectural domain), dimensions and material choice can explain mass (given that dimensions imply volume 95 and material choice implies density). However, these causal relationships may not be relevant to the designer's efforts to mitigate the risk of volatile variables and variable change propagation across the system. Furthermore, explanatory variables cannot be assumed to be independent variables in the engineering sense in which designers can directly control their values. An example of such a variable may include oil prices-which exist in the specification domain and are outside the direct control of the designer. Perhaps the biggest challenge in advancing this econometrics-inspired framework is selecting which cases of past PD projects to compare in a portfolio, given that variables that are explanatory for certain products may not be so for others. For instance, in the RWW example explained in Section 8.1, we argue that for the typical profit-maximizing firm, market-oriented constraints take precedence over manufacturing-oriented constraints. However, in footnote 20, we explain how some profit-maximizing firms may not be concerned with the product produced necessarily turning a profit if it is within the context of a larger corporate strategy. Consider the Bugatti Veyron, the Volkswagen (VW) Group's claim to producing the world's fastest production car. Despite leading to billions in financial losses while producing the vehicle for over a decade, the Veyron earned the VW Group prestige within the auto industry as a technological achievement ("Zoom, Sputter, Aagghhh!!" 2013). Thus, it is clear that many of the variables pertaining to market demand were not explanatory, and accordingly constraints pertaining to manufacturing attained much higher emphasis than those pertaining to market. It is important to note that the case-dependence of whether certain variables are explanatory limits the generalizability of knowledge somewhat from this exercise. In the short run, the critical caveat to this approach is that variables found to be explanatory among a portfolio of similar products is expected to be explanatory for other similar products and applications, only. However, as I point out in the conclusion, this caveat may become less relevant as more cases are studied. 8.3 Conclusion Noting that constraints undergo a gradual process of activation on the specification domain, the previous sections identify the potential usefulness of defining a preferred order for constraint propagation in the PD process. We propose an econometrics-inspired framework for addressing 96 this question in which explanatory and response variables from past PD projects can potentially be isolated to infer causal relationships. With this knowledge, designers may begin to identify an order in which certain constraints should be activated over others for projects that are similar to the ones studied. Despite the initial caveat of case-dependence, as more portfolios of previous projects are studied, it is conceivable that trends may be identified, and thus categories that are broader than a particular class of products can be identified for which heuristics regarding ordered constraint propagation can be made. In other words, as more cases are studied, the likelihood that knowledge may be generalized and that preferred orders for constraint propagation may be identified for broad applications beyond simply similar products increases. 97 9. Bringing it Together: How the Framework Influences the Way We Think about Design Given the cumulative nature in which the framework is presented and discussed in this work, the purpose of this chapter, acting in lieu of a traditional concluding chapter for a thesis, is to collect and summarize some of the big picture ideas to emerge from reasoning with framework starting from Chapter 3. Thus, this chapter is intended to review potential contributions of this work in light of our pursuit of a deeper understanding of the PD process. In Section 9.1, I discuss what the framework as presented in this thesis hopes to achieve. Following this, in Section 9.2, I summarize some of the new ways of thinking about the PD process to emerge from developing this framework. In Section 9.3, I identify some areas for future work that can expand upon and give greater meaning to the framework as presented in this thesis. Finally, I offer some concluding thoughts in Section 9.4. 9.1 What Does This Work Hope to Achieve? In Section 3.1, I summarized the difficulties in advancing our work on Set-Based Thinking due to a lack of standardized terminology and a subsequent breakdown in communication and understanding when discussing certain elements of the PD process. Contextualizing this observation in a broader call for common terminology and a common model for the design process from Blessing, we defined the terms for a novel framework for product development. First-order, this framework provides a new qualitative description of the PD process. However, in doing so, it also: 1. Re-conceptualizes how we describe the front end of the PD process 2. Challenges implied or assumed causal structures or relationships in the PD process 21 21 This statement should not be misconstrued as a claim to have an understanding of how things actually work in the PD process; rather, I present a different take on certain elements of the process that questions things that are often assumed to be true in the literature or in general discussion about the design process. 98 3. Presents groundwork for investigating latent research challenges in our community Exactly how we achieve these three points shall be made clear in the following section. Each of the points above, 1-3, shall be mapped specifically to the new ways of thinking that are described in Section 9.2 in Table 3. 9.2 What New Ways of Thinking Are Proposed? Articulating a novel framework for PD that is the focus of this work since Chapter 3 has yielded a variety of fascinating insights about the fundamentals of the process and the way we design. I collect major insights as they are established, roughly chronologically, throughout the thesis and summarize them in each of the subsections below. At the end of this section, I map these insights and subsections to the three points in Section 9.1 thus classifying the underlying goals of this work. Set-Based Thinking as a Way to Enable the Consideration of More Options In Chapter 2, across a broad sampling of PD-affiliated literature, I identify characteristics of processes that have been discussed and studied as good PD practices since the aforementioned Ward et al. 1995 case studies on Toyota. I then use these characteristics to identify the Principles of Set-Based Thinking, a new way of categorizing a set of similarly-motivated techniques and methodologies which transcend our original starting point of studying the SBD literature. Careful consideration of these Principles allows us to assert that the perceived strength of a Set-Based Thinking approach is that it enables the consideration of more options. Recognizing that the consideration of more options is a well-respected heuristic in the design literature, we take a step back to raise questions about its practical effectiveness as a strategy in light of the lack of novel outcomes attained by some practitioners of SBD. Trying to understand what it actually means to consider more options while proposing standardized terminology that was inspired by elements of the Set-Based Thinking literature was an excellent frame of reference by which to begin the development of the novel framework for PD, which began in earnest in Chapter 3. After carefully constructing the framework over Chapters 3-7, we later find in Section 7.4 that novel 99 outcomes, in fact, cannot be guaranteed, even if a large raw number of options are considered. The inescapable limits of bounded rationality, as we summarize below, inherently limit the quality and quantity of options that we can simultaneously consider, which is why attaining novel outcomes become a chance pursuit. A Temporal Description of the Process via the Activation of Constraints Shortly after describing the process using the terms I introduced in Chapter 3, I establish a fundamental tenet upon which the rest of the framework builds-that the PD process is temporally driven by the propagation or activation of dormant constraints. Given that we have approached describing the process from a set-based perspective as discussed in Chapter 2, it should naturally follow that in order to move the process forward, constraints must be applied to the initial set of possibilities, eventually narrowing it to the selection of a single point (i.e. the production version of a design). Thus, rather than proposing a divergent-convergent process that is typical of most of the PD literature, the framework describes an idealized process of gradual convergence in order to select the final product design. While convergence over a broad initial set of possibilities is not a new notion to the SBD and related literature, the framework's contribution is to formalize this description via its carefully defined terms so as to promote common understanding and to guide deeper insight into the mechanics of the process. Using the richer language of the framework, instead of generalizing this process as "gradual convergence," we can be more precise in our description, for instance stating that "as constraints are slowly made active over the initial feasible set in the specification domain, resolution on architectures explored increases while the set narrows." Of course, from this richer description of the process, we may ask: how is the first subset of constraints gradually activated? Our answer, as we shall address in the next subsection, is the articulation of form. Re-Conceptualizing the Relationship between Function and Form in Early-Stage Design Most current theories for PD imply that function maps to form-however, I disagree with this assertion when describing the front end of the process. While function motivates form, I assert 100 that "function mapping to form" is not the appropriate causal relationship to explain the mechanics of early stage design. The view espoused by the framework is most succinctly stated by the Surjective Theorem (introduced in Section 5.1) which implies that the architectural domain maps to the specification domain-or that form maps to function. This assertion can be approached from a variety of perspectives from smaller, cumulative insights as they are made throughout the thesis. A few of these insights are summarized below: " The framework's definition for architecture: As established in Section 3.4, unlike most existing design theories, our definition of architecture does not try to formalize the relationship between function and form. This is established from our understanding of the relationship between the two domains in which we work. - The specification domain is an artificial construct: The specification domain is where we measure the success of our design in carrying out its intended functions. It is also an artificial construct that is created by designers to organize our thinking about something that, unlike form, is not innately intuitive (i.e. product function). Thus, the only way to explore this domain is to map architectures onto it. " How PD in practice differs from the idealized model presented by the framework: As addressed in Chapter 6, deviations from our ideal take on the PD process, such as a lack of knowledge of the surjective function, the uncertainty of variables, and the unknown bounds of feasibility are managed via the expression of form. The result of these observations is that the articulation of form precedes (and in fact, enables) the exploration of function. While "zig-zagging" (Suh 1995) between function and form is common in practice (i.e. one form is tested and measured for its function, which then informs the selection of the next form to try out, etc.), it is somewhat misleading to imply that function maps to form when the first step to exploring function is to propose some form of arbitrary resolution. This perspective bears great significance in discussing the Designer's Dilemma, addressed below. The Designer's Dilemma and its Relation to Fixation The relationship of form onto function implies that we cannot directly sample the specification domain in early stage design (by analytical means or otherwise). This fact, coupled 101 with our innate human cognitive limitations as expressed by bounded rationality, implies that as we continue to generate ideas, we are inherently influenced by the nature of past ideas explored. Furthermore, analyzing the mechanisms underlying ideation in early stage design, we take a strong view on fixation-concluding that defixation strategies that are based on "indiscriminate search" do not prevent the Designer's Dilemma as they do not fundamentally alter the flawed 22 way in which we currently propose ideas. Of course, describing concept development and ideation particularly in early-stage design as "indiscriminate" or "blind" search is a result from our analysis using the Designer's Dilemma in Chapter 7. It is important to note, as we do earlier, that this search is not truly "random," as a random search would not be affected by fixation and may in fact produce a higher variance of ideas (and thus greater diversity of options considered) in practice. In light of this, we propose a distinction between the "blind" search that is typical of the vast majority of current ideation techniques and methods for "informed search," which is identified as an area for future work. However, without fundamental improvements to our current methods of "indiscriminate" or "blind" search, novel outcomes are largely up to chance and cannot be guaranteed. Proposing a Preferred Order for Constraint Propagation Given the central role of constraints in moving the process forward, we assert the existence of a preferred order by which constraints should be propagated in order to avoid the deleterious consequences of rework, change propagation, and similar problems that are often encountered during the development of our increasingly complex product development systems. In other words, our hypothesis is that improper constraint propagation whereby certain volatile parameters are prematurely locked down can expose our systems to the unanticipated consequences of changes that inevitably occur due to the high levels of uncertainty during the PD process. While we do not have the luxury in this work to build and test such a framework, we propose the foundation for determining a preferred order for constraint propagation using 22 The process is inherently flawed since we are forced to articulate form within the limits of our bounded rationality and inability to simultaneously compare many alternatives at once. 102 econometric-modeling inspired approach (see Chapter 8). With an understanding of the explanatory variables for a given design problem, we can begin the process of inferring which constraints should be activated prior to others based on our appreciation of pertinent causal relationships and ensuing variable sensitives. As we discuss in Chapter 8, a caveat to this approach is that the results can only generalize to similar problems explored; however, as more examples are studied, the potential for finding and establishing more universal heuristics certainly increases. Reinterpretations of Existing Work on Design Theory During our development of the framework, we have also come across a few opportunities to reinterpret various facets of established design theory, in many cases giving new meaning to these streams of work. A few these reinterpretations are summarized below, along with references to where detailed discussion on these points can be found in the text: - Why Set-Based Thinking cannot guarantee novel outcomes: As we establish during our discussion of the Designer's Dilemma, unanticipated outcomes cannot be expected from following an SBD approach. This is discussed in further detail in Section 7.4. - MDO, Pareto efficiency, and their limitations in the front end of the PD process: The framework presents an idealization of PD whereby the entire process can be thought of optimizing in the specification domain, detailed in Chapter 5. However, these idealizations do not apply to the front end with which we are particularly interested in this thesis, as addressed in Chapter 6.2. - The Pareto frontier as a subset of the surjection between the architectural and specification domains: This observation is first addressed in Section 5.3, and is also elaborated upon further in the Appendix. - Why artificial intelligence was and remains a difficult pursuit for our research community: Without an understanding of how to analogize human cognitive processes during early stage design into a reliable algorithm, the framework instructs us that the abstract threat of artificial intelligence replacing or subsuming high-level human decisionmaking remains hypothetical and not actionable (further discussion is found in Section 7.4). This observation also explains the more realistic outlook the design theory community has adopted for this line of work over the years. 103 Interpreting the Contributions of this Work In Section 9.1, we describe how the framework hopes to provide a new qualitative description of this process, in addition to three additional underlying goals. These goals are mapped to the new ways of thinking that have been summarized in Table 3. Consider the example of a temporal description of the process via the activation of constraints. While the description of constraints transitioning from "dormant" to "active" states is a re-conceptualization of the front end, it does not challenge any implied causal relationships that are assumed to be true in the engineering design literature. However, regarding re-conceptualizing the relationship between function for form, not only does this repaint a picture of activity in the front end, but it also challenges the assumption of function onto form. Table 3: Interpreting the Contributions of this Work What Does This Work Hope to Achieve? What New Ways of Thinking are Proposed? Re-Conceptualizes Front End Challenges Assumed Causal Relationships Presents Groundwork for Investigating Latent Research Challenges Set-Based Thinking and Considering More Options Temporal Description of the Process via Constraint Activation X Re-Conceptualizing Relationship between Function and Form X X The Designer's Dilemma X Preferred Order for Constraint Propagation Reinterpreting Existing Design Theory 104 X 9.3 What are Areas for Future Work? In our discussion over the previous section, we have alluded to a number of areas for future work. These topics, in addition to a few which thus far have not been formally addressed in the text, are discussed in the subsections to follow. A Call for a Deeper Cognitive Understanding of the Front End of the Process Central to building up to the Designer's Dilemma and analyzing the ensuing issues is the need to pursue of a deeper understanding of the mechanics underlying cognition during the creative pursuit of early stage design. Building such knowledge may reveal the key to preventing the inherent fixation that results from the Dilemma. Pursuant to studying this are a few potential areas of inquiry: " Developing techniques to sample the limits of feasibility and enable "informed search": Given that the pursuit of optimality often occurs at the bounds of feasibility, techniques to enable the search of these bounds in early-stage design may bear promise for more robust concept generation methods. " Exploring cross-domain variable sensitivity: Understanding how changes in design parameters in the architectural domain map to changes in specifications in the specification domain may help enable more efficient ideation to sample larger portions of the solution space as desired. - What cognitive mechanisms are at play during common concept development and opportunity identification exercises?: Answering this question can help us objectively analyze how effective certain ideation techniques are (e.g. gallery method, morphological matrix, etc.). With this knowledge, can these ideation techniques be redesigned to more effectively sample the feasible set of options? Developing an Econometrics Approach to PD Modeling As referenced in Chapter 8 and again in Section 9.2, the use of regression analysis techniques from econometrics can potentially yield useful insights into proposing a preferred order 105 for constraint propagation for certain PD projects. However, these modeling techniques bear great potential to address questions beyond the issue of constraint propagation. Given the simultaneous pressures of 1) the increasing complexity of the products we design (Lewis 2012), and 2) the intricate broader social, political, and economic contexts in which consumers can no longer be regarded passive end users of the products we create (von Hippel 2005), it is largely inadequate to apply engineering domain knowledge that is not at least informed by. expertise from other areas to our problems. Due to the current difficulties we encounter in accounting for external, traditionally non-technical factors within our engineering-driven models and methodologies for PD, it may be of significant benefit to the design literature to incorporate more rigorous research methods from the social sciences to our models, such as econometric methods, particularly for early stage design. The potential implications are broad and fascinating. For instance, understanding causal relationships within our PD systems (among factors that are not strictly technical) may help designers transcend the design of products in the short-term, enabling them to predict and proactively design for nascent market needs. Expanding the Framework In Chapter 1, we describe an analogy to the development of our framework to that of common law. In this sense, the framework is continually emergent, expanding it to address more questions about the PD process will only bear greater meaning to prior established elements and principles. Thus, there are many opportunities to expand the framework established thus far, a non-exhaustive list of which is presented below: - Expanding Set-Based Thinking: In (Ghosh and Seering 2014), we identify a number of potential inquiries to expand the Set-Based Thinking paradigm described in Chapter 2. While the remainder of the thesis does not necessarily address these project managementoriented issues, addressing questions such as "which types of organizations are more suited for SBD?" and "what is the cost of pursuing SBD and is it always worth the cost?" are particularly pertinent to practitioners who wish to apply SBD and affiliated methods. " Addressing system design concerns: Designing multi-level PD systems pose a unique set of challenges for designers, who must consider a host of issues such as those associated with the integration of individual subsystems into the greater system (Blanchard and 106 Fabrycky 1990). While the examples discussed thus far in the framework have not necessarily been described from a multi-level system perspective, there is no reason to believe that subsystem and integration design tasks cannot be represented within this framework. Developing examples and exploring how, for example, integration tasks can be accounted for (perhaps via added dormant constraints) is a topic of future study. Formally incorporating objectives: Thus far, we have accounted for the exploration of product function via the articulation and measurement of specifications. Given the analogy of the PD process to conducting an optimization problem in the specification domain, it would be appropriate to formally incorporate notions of an objective function into the framework.23 Currently this information is handled indirectly via the propagation of constraints (for example, those pertaining to consumer demand), which can theoretically be expressed analytically in the specification domain. 9.4 Concluding Thoughts - "A science of artificialphenomena is always in imminent danger of dissolving and vanishing." Hebert A. Simon, The Sciences of the Artificial In this thesis, I develop and study a novel framework for PD. While the framework as described in this work is still quite nascent, it is my sincere hope that our discussion thus far will have caused the reader to think about the seemingly familiar PD process in new ways. If I am extremely fortunate, this framework will have inspired the reader to pursue questions that, due to my fixation with the task of articulating this thesis, I would not have anticipated myself. Even if 23 I do not anticipate elements of the global objective function to be mutually exclusive to the expression of the surjection which maps architecture to specification. However, whereas the surjection is the analytic function which accepts design parameters as inputs and provides specifications as outputs, the objective function takes inputs of specifications (and design parameters indirectly) and outputs aggregated objective values (values which weigh the influence of individual specifications in a design, for instance). Thus, these concepts are distinct, where the specification domain could potentially be thought to map to another space which describes design objective value. 107 I am not that fortunate, I would hope that this thesis will have convinced the reader that the fundamentals of the PD process are still worth studying. I opened this work with the context of Simon's plea for a "teachable doctrine" for design. To teach is to inspire, and even if this framework does not embody the full intent of Simon's vision for a teachable doctrine, if the reader leaves with an appreciation for studying the enigmatic yet worthwhile challenge that is the product development process, then I have succeeded at my task. 108 References Antonsson, E. K., and K. N. Otto. 1995. "Imprecision in Engineering Design." Journal of MechanicalDesign117 (B): 25-32. "Artificial General Intelligence." Artificialgeneralintelligence. 2015. Wikipedia. https://en.wikipedia.org/wiki/ Bilton, Nick. 2015. "Disruptions: On the Fast Track to Routine 3-D Printing." New York Times. http://bits.blogs.nytimes.com/2013/02/17/disruptions-3-d-printing-is-on-the-fasttrack/?nl=todaysheadlines&emc=editth_20130218&_r=0. Blanchard, B. S., and W.J. Fabrycky. 1990. Systems EngineeringandAnalysis. Englewood Cliffs, N.J.: Prentice Hall. Blessing, Lucienne. 2003. "Future Issues in Design Research." In Human Behaviour in Design, edited by Udo Lindemann, 298-303. Berlin: Springer-Verlag Berlin Heidleberg. "BMW 3 Series (E90) 2005-2008." 2015. Autoevolution. http://www.autoevolution.com/ cars/bmw-3-series-e90-2005.html#aengbmw-3-series-sedan-2005-318i. "BMW 3 Series (E90) 2008-2011." 2015. Autoevolution. http://www.autoevolution.com/ cars/bmw-3-series-e90-2008.html#. Braha, Dan, David C Brown, Amaresh Chakrabarti, Andy Dong, Georges Fadel, Jonathan R A Maier, Warren Seering, David G Ullman, and Kristin Wood. 2013. "DTM at 25: Essays on Themes and Future Directions." In ASME International Design Engineering Technical Conferences. Portland, OR. Canbaz, Baris, Bernard Yannou, and Pierre-Alain Yvars. 2013. "Expanding the Bottom-Up Approach Through Intergrating Design Attitudes into Set-Based Design." In ASME InternationalDesign EngineeringTechnical Conferences. Portland, OR. "Cardinality of Surjection." CardinalityofSurjection. 2015. Proof Wiki. https://proofwiki.org/wiki/ Carlos, S., Kaarthic Madhavan, G. Gupta, A. Keese, U. Maheshwaraa, and Carolyn C Seepersad. 2006. "A Flexibility-Based Approach to Collaboration in Multiscale Design." In 11th AIAA/ISSMO MultidisciplinaryAnalysis and Optimization Conference. Portsmouth, VA. Chanron, Vincent, and Kemper Lewis. 2005. "A Study of Convergence in Decentralized Design Processes." Research in EngineeringDesign 16 (3): 133-145. 109 "Chinese Room." 2015. Wikigedia. https://en.wikipedia.org/wiki/Chineseroom. Cormier, Phillip, Andrew Olewnik, and Kemper Lewis. 2013. "Towards a Formalization of Affordance Modeling in the Early Stages of Design." In ASME International Design Engineering TechnicalConferences. Portland, OR. Dahmus, Jeffrey B, Javier P Gonzalez-Zugasti, and Kevin N Otto. 2000. "Modular Product Architecture." In ASME International Design Engineering Technical Conferences. Baltimore, MD. Day, George S. 2007. "Is It Real? Can We Win? Is It Worth Doing?" HarvardBusiness Review 85 (12): 110-120. De Neufville, Richard. 2003. "Real Options: Dealing with Uncertainty in Systems Planning and Design." IntegratedAssessment 4 (1): 26-34. Devendorf, Erich, and Kemper Lewis. 2011. "The Impact of Process Architecture on Equilibrium Stability in Distributed Design." JournalofMechanicalDesign133 (10): 1-12. Dym, Clive L., Alice M. Agogino, Ozgur Eris, Daniel D. Frey, and Larry L. Leifer. 2005. "Engineering Design Thinking, Teaching, and Learning." Journal ofEngineeringEducation 94 (1): 103-120. Finch, William W, and Allen C Ward. 1997. "A Set-Based System for Eliminating Infeasible Designs in Engineering Problems Dominated by Uncertainty." In ASMEInternationalDesign EngineeringTechnical Conferences. Sacremento, CA. Finger, Susan, and John R. Dixon. 1989. "A Review of Research in Mechanical Engineering Design. Part I: Descriptive, Prescriptive, and Computer-Based Models of Design Processes." Research in EngineeringDesign (1): 51-67. "Fixation." 2015. Dictionary.com Unabridged Accessed July 5. http://dictionary.reference.com/ browse/fixation. Ford, D.N., and D.K. Sobek. 2005. "Adapting Real Options to New Product Development by Modeling the Second Toyota Paradox." IEEE Transactionson EngineeringManagement52 (2): 175-185. "Fourteenth Amendent of the United States Constitution." 2015. Wikipedia. https://en.wikipedia.org/wiki/FourteenthAmendmenttotheUnitedStatesConstitution. Friedman, Thomas L. 2015. "Hillary, Jeb, Facebook and Disorder." New York Times. http://www.nytimes.com/2015/05/20/opinion/thomas-friedman-hillary-jeb-facebook-anddisorder.html?_r-0. 110 Galvan, Edgar, and Richard Malak. 2012. "A Genetic Algorithm Approach for Technology Characterization." In ASME International Design Engineering Technical Conferences. Chicago, IL. Ghosh, Sourobh, Erich Devendorf, and Kemper Lewis. 2014. "Testing the Parallel Character Hypothesis for Distributed Design Procceses Subjected to Stochastic Disruptions." Artificial Intelligence for EngineeringDesign, Analysis & Manufacturing28 (4): 399-412. Ghosh, Sourobh, and Warren Seering. 2014. "Set-Based Thinking in the Engineering Design Community and Beyond." In ASME International Design Engineering Technical Conferences. Buffalo, NY. Gil, Nuno, Sara Beckman, and Iris D Tommelein. 2008. "Upstream Problem Solving Under Uncertainty and Ambiguity: Evidence From Airport Expansion Projects." IEEE Transactions on EngineeringManagement55 (3): 508-522. Goodwin, Tom. 2015. "The Battle Is for the Customer Interface." TechCrunch. http://techcrunch.com/2015/03/03/in-the-age-of-disintermediation-the-battle-is-all-for-thecustomer-interface/#.derbth:9SFc. Gossard, David. 2009. "Sketching and Drawing." MIT OpenCourseWare. http://ocw.mit.edu/courses/mechanical-engineering/2-007-design-and-manufacturing-ispring-2009/lecture-notes/MIT2_007s09_lecO4.pdf. Haggman, Anders, Tomonori Honda, and Maria C Yang. 2013. "The Influence of Timing in Exploratory Prototyping and Other Activities in Design Projects." In ASME International DesignEngineeringTechnical Conferences. Portland, OR. Helms, S, F G H Behncke, L Lindl6f, M C Wickel, and U Lindemann. 2014. "Classification of Methods for the Indication of Change Propagation - A Literature Review." In International Design Conference - DESIGN2014, 211-220. "HSC Dolphin Jet." 2013. http://en.wikipedia.org/wiki/HSCDolphinJet. Hubka, Vladimir, M. Myrup Andreasen, and W. Ernst Eder. 1988. PracticalStudies in Systematic Design. London: Butterworths. Iansiti, Marco. 1995. "Shooting the Rapids: Managing Product Development in Turbulent Environments." CaliforniaManagementReview 38 (I): 37-58. Jansson, DG, and SM Smith. 1991. "Design Fixation." Design Studies 12 (1): 3-11. 111 Jiao, Jianxin, Timothy W. Simpson, and Zahed Siddique. 2007. "Product Family Design and Platform-Based Product Development: A State-of-the-Art Review." Journal of Intelligent Manufacturing18 (1): 5-29. Kelly, Danny. 2013. "The Advantages and Disadvantages of the No-Huddle, Hurry-up Offense in the NFL." SB Nation. http://www.sbnation.com/2013/9/13/4722170/no-huddle-hurry-upoffense-nfl-peyton-manning-chip-kelly. Kerga, Endris. 2013. "Set-Based Concurrent Engineering: A Learning Method to Increase Awamess Level in Industry & A Methodolgy to Identify and Prioritize Areas at a Product Leve." Ph.D. thesis, Politecnico di Milano. Lee, Seung-Il, Jun-Seo Bae, and Young Sang Cho. 2012. "Efficiency Analysis of Set-Based Design with Structural Building Information Modeling (S-BIM) on High-Rise Building Structures." Automation in Construction23 (May): 20-32. Lewis, Kemper. 2012. "Making Sense of Elegant Complexity in Design." JournalofMechanical Design 134 (12): 120801. Lewis, Kemper, and Farrokh Mistree. 1998. "Collaborative, Sequential, and Isolated Decisions in Design." JournalofMechanicalDesign120 (12): 643-652. Linsey, J. S., I. Tseng, K. Fu, J. Cagan, K. L. Wood, and C. Schunn. 2010. "A Study of Design Fixation, Its Mitigation and Perception in Engineering Design Faculty." Journal of MechanicalDesign 132 (4): 1-12. Madhavan, Kaarthic. 2007. "A Framework for a Flexibility- Based Approach to Multiscale and Multidisciplinary Design." M.S thesis, University of Texas at Austin. Madhavan, Kaarthic, David Shahan, Carolyn C Seepersad, Danny A Hlavinka, and Walt Benson. 2008. "An Industrial Trial of a Set-Based Approach to Collaborative Design." In ASME InternationalDesign Engineering Technical Conferences. New York, NY. Maier, Jonathan R. a., and Georges M. Fadel. 2008. "Affordance Based Design: A Relational Theory for Design." Research in EngineeringDesign 20 (1) (December 12): 13-27. Malak, Richard J., Jason Matthew Aughenbaugh, and Christiaan J.J. Paredis. 2009. "MultiAttribute Utility Analysis in Set-Based Conceptual Design." Computer-AidedDesign 41 (3) (March): 214-227. Malak, Richard J., and Christiaan J. J. Paredis. 2010. "Using Parameterized Pareto Sets to Model Design Concepts." JournalofMechanicalDesign132 (4): 1-11. 112 Mebane, Walter L., Craig M. Carlson, Chris Dowd, David J. Singer, and Michael E. Buckley. 2011. "Set-Based Design and the Ship to Shore Connector." NavalEngineersJournal123 (3): 79-92. Meyer, Marc H., and Alvin P. Lehnerd. 1997. The Power ofProductPlatforms. New York: Free Press. Nahm, Y.-E., and H. Ishikawa. 2006. "A New 3D-CAD System for Set-Based Parametric Design." The InternationalJournalofAdvanced ManufacturingTechnology 29 (1-2) (April 5): 137150. Nykamp, D.Q. "Countably Infinite definition/countablyinfinite. Defintion." Math Insight. http://mathinsight.org/ Oberg, J. 1999. "Why the Mars Probe Went off Course." IEEE Spectrum Magazine. Oosterwal, Dantar. 2010. The Lean Machine:HowfHarley-DavidsonDrove Top-Line Growth and Profitabilitywith RevolutionaryLean ProductDevelopment. New York: AMACOM. Orsborn, Seth, Jonathan Cagan, Richard Pawlicki, and Randall C. Smith. 2006. "Creating Crossover Vehicles: Defining and Combining Vehicle Classes Using Shape Grammars." Artificial Intelligence for EngineeringDesign, Analysis and Manufacturing20 (03) (June 27): 217246. Pahl, Gerhard, and Wolfgang Beitz. 1988. EngineeringDesign:A Systematic Approach. London: The Design Council. Pahl, Gerhard, Wolfgang Beitz, Jorg Feldhusen, and Karl-Heinrich Grote. 2007. Engineering Design:A Systematic Approach. 3rd ed. London: Springer. Panchal, Jitesh H., Marco Gero Fernandez, Christiaan J.J. Paredis, Janet K. Allen, and Farrokh Mistree. 2007. "An Interval-Based Constraint Satisfaction (IBCS) Method for Decentralized, Collaborative Multifunctional Design." ConcurrentEngineering15 (3): 309-323. "Pareto Efficiency." 2015. Paretoefficiency#Paretofrontier. Wikipedia. https://en.wikipedia.org/wiki/ Parker, Robert, and Richard J. Malak. 2011. "Technology Characterization Models and Their Use in Designing Complex Systems." In ASME InternationalDesign Engineering Technical Conferences. Washington, DC. Parrish, Kristen, John-michael Wong, Iris D Tommelein, and Bozidar Stojadinovic. 2007. "Exploration of Set-Based Design for Reinforced Concrete Structures." In Proc. of15th Conf of the Int. Group for Lean Construction, 213-222. Michigan, USA. 113 Poppendieck, Mary, and Tom Poppendieck. 2003. Lean Software Development:An Agile Toolkit. Boston, MA: Addison-Wesley Professional. Pugh, Stuart. 1991. Total Design: IntegratedMethods for Successful Product Engineering. New York: Addison-Wesley. Raudberget, Dag. 2010. "Practical Applications of Set-Based Concurrent Engineering in Industry." JournalofMechanical Engineering56 (11): 685-695. Robertson, David, and Karl Ulrich. 1998. "Planning for Product Platforms." Sloan Management Review 39 (4): 19-3 1. Sacks, Rafael, Lauri Koskela, Bhargav A Dave, and Robert Owen. 2010. "Interaction of Lean and Building Information Modeling in Construction." Journalof Construction Engineeringand Management 136 (9): 968-980. Schmidt, Elisabeth. 2015. "An Approach for Structuring Methods in Product Development Processes and the Role of Uncertainties." M.S. thesis, Technische Universitat M-inchen. Schwaber, Ken, and Jeff Sutherland. 2013. "The Scrum Guide TMi" "Scrum Primer." http://www.infoq.com/minibooks/ScrumPrimer. Seepersad, Carolyn C, David Shahan, and Kaarthic Madhavan. 2007. "Multiscale Design for Solid Freeform Fabrication." In Solid Freeform FabricationSymposium. Austin, TX. Shah, Jami J., Noe Vargas-Hernandez, Joshua D. Summers, and Santosh Kulkarni. 2001. "Collaborative Sketching (C-Sketch) - An Idea Generation Technique for Engineering Design." Journalof CreativeBehavior35 (3): 168-198. Shan, S., and G.G. Wang. 2004. "Space Exploration and Global Optimization for Computationally Intensive Design Problems: A Rough Set Based Approach." StructuralandMultidisciplinary Optimization28 (6) (August 24): 427-441. "Ship-to-Shore Connector Implementation of Set-Based Design." 2009. Internal Report. NAVSEA, Washington, D.C. Simon, Herbert A. 1996. The Sciences ofthe Artificial. 3rd ed. Cambridge, MA: MIT Press. Simpson, Timothy W. 2004. "Product Platform Design and Customization: Status and Promise." AI EDAM: Artificial Intelligence for EngineeringDesign, Analysis and Manufacturing 18 (01): 3-20. 114 Singer, David J., Norbert Doerry, and Michael E. Buckley. 2009. "What Is Set-Based Design?" Naval EngineersJourna121 (4) (October 11): 31-43. Sobek, Durward K., Allen C Ward, and Jeffrey K Liker. 1999. "Toyota's Principles of Set-Based Concurrent Engineering." Sloan ManagementReview 40 (2): 67-83. "Stare Decisis." 2015. Legal Information https://www.law.cornell.edu/wex/staredecisis. Institute. Accessed July 14. Stiny, G. 1980. "Introduction to Shape and Shape Grammars." Environment and Planning 7 (November): 343-351. Suh, Nam P. 1995. "Axiomatic Design of Mechancial Systems." Journal ofMechanicalDesign 117 (6): 2-10. "Superintelligence." 2015. Wikipedia. https://en.wikipedia.org/wiki/Superintelligence. "Surjective Function." 2015. Wikipedia. https://en.wikipedia.org/wiki/Surjectivefunction. Sutherland, Jeff, Anton Viktorov, Jack Blount, and Nikolai Puntikov. 2007. "Distributed Scrum: Agile Project Management with Outsourced Development Teams." In Proc. of the 40th Hawail InternationalConference on System Sciences, 1-10. Thomke, Stefan. 2003. Experimentation Matters. 1st ed. Boston, MA: Harvard Business School Publishing Corporation. Thomke, Stefan, and Donald Reinertsen. 1998. "Agile Product Development: Managing Development Flexibility in Uncertain Environments." CaliforniaManagementReview41 (I): 8-30. Ulrich, Karl T. 1994. "The Role of Product Architecture in the Manufactruing Firm." Research Policy24 (3): 419-440. Ulrich, Karl T., and Steven D. Eppinger. 2012. ProductDesign and Development. 5th ed. New York: McGraw-Hill, Inc. Viswanathan, Vimal K., and Julie S. Linsey. 2013. "Role of Sunk Cost in Engineering Idea Generation: An Experimental Investigation." JournalofMechanicalDesign135 (12): 1-12. Von Hippel, Eric. 2005. DemocratizingInnovation. Cambridge, MA: MIT Press. Ward, Allen C. 1989. "A Theory of Quantitative Inference for Artifact Sets, Applied to a Mechanical Design Compiler." Massachusetts Institute of Technology. 115 Ward, Allen C, and Warren Seering. 1989. "Quantitative Inference in a Mechanical Design 'Compiler."' In ASME InternationalDesign EngineeringTechnical Conferences. Ward, Allen, Jeffrey K Liker, John J Cristiano, and Durward K. Sobek. 1995. "The Second Toyota Paradox: How Delaying Decisions Can Make Better Cars Faster." Sloan Management Review 36 (3): 43-61. Weisstein, Eric W. 2015. "Surjection." Wolphram Math World http://mathworld.wolfram.com/ Surjection.html. Womack, James P., Daniel T. Jones, and Daniel Roos. 2007. The Machine That Changed the World New York. Yang, Maria C. 2005. "A Study of Prototypes, Design Activity, and Design Outcome." Design doi:10.1016/j.destud.2005.04.005. 649-669. (November): 26 (6) Studies http://linkinghub.elsevier.com/retrieve/pii/S0142694X050003OX. Yang, Maria C. 2008. "Observations on Concept Generation and Sketching in Engineering Design." Research in EngineeringDesign20 (1): 1-11. Yi, S., J. Shin, and G. Park. 2008. "Comparison of MDO Methods with Mathematical Examples." StructuralandMultidisciplinaryOptimization3 5 (3): 391-402. "Zoom, Sputter, Aagghhh!!" 2013. The Economist. http://www.economist.com/blogs/ graphicdetail/2013/09/daily-chart-i 8?fsrc=scn/fb/wl/tr/zoomsputteraagghhh. 116 Appendix Herein we review a handful of topics for further edification for the reader. They represent ideas which I could not formally validate during the duration of the thesis, and thus are referenced here in order to be studied and formalized in future work. Al The Pareto Frontier as a Subset of the Surjection's Codomain In Section 5.3, we assert the possibility that the Pareto frontier is a subset of the codomain of the analytic surjective function as expressed in this work. Let us begin with a mathematical description of the Pareto frontier, P(Y), provided from Wikipedia ("Pareto Efficiency" 2015). We have a system with function f: R' -> R', where X is a compact set of feasible decisions in the metric space R', and Y is the feasible set of criterion vectors in R' such that Y = fy E Rm: y = f(x), x E X}. Assuming the preferred directions of criteria values are known, a point y" E Rm is preferred to another point y' E R:. Thus, the Pareto frontier is written as: P(Y) = fy' E Y: fy" E Y y" > y', y" # y'} = 01. Our focus in this definition is on the function fwhich maps points in the set X to the set Y. This function f in the definition above maintains that for every y in Y, there must be an element x in X such that f (x) = y. Fundamentally, this is a re-articulation of the definition for a surjection (see Section 5.1). Thus, we claim that f is in fact a surjection, and that P(Y) is a subset of the codomain of f (i.e. the set Y). A2 Cardinality of the Two Domains Another property of a surjective function is the cardinality of its domain is greater than or equal to the cardinality of its codomain. A proof can be found in the following reference ("Cardinality of Surjection" 2015). 117 With the Surjective Theorem, at a fixed level of resolution, we can assert that the number of instances in the architectural domain is equal to or greater than or equal to the number of instances in the specification domain. Intuitively, this should follow from our understanding that each instance in the specification domain must have at least one instance from the architectural domain mapping to it (and that as a mathematical function, one instance in the architectural domain cannot map to more than one instance in the specification domain). Suppose we knew of all the instances in the architectural and specification domains, respectively, and that we could count them. Now suppose we are given the task to randomly choose two instances from the specification domain, which we shall call pair A. For simplicity, let us assume that there is only one specification of interest in the specification domain. After selecting pair A, we measure their difference in specificity values. Now, putting pair A back into the specification domain, we are asked to choose two instances at random from the architectural domain. We then map these two instances from the architectural domain to their corresponding instances in the specification domain. Let us call these corresponding instances in the specification domain pair B. As before, we measure the difference in specification values between pair B. Now we are asked: is the difference between the specification values in pair B as likely to be as great as it is in pair A? Given the cardinality of the two domains, I assert that pair B (the pair of instances we mapped from those that we selected in the architectural domain) are not as likely to be as different as pair A (the pair that we directly selected from the specification domain). This property bears interesting implications for how designers ideate. For instance, if we chose two instances at random from the architectural domain, even if these instances have diverse design parameter values, will these differences map to proportional differences in the specification values in the specification domain? That is, even if we make the effort to articulate two architectures that are seemingly different from one another, is the performance of these architectures likely to be as different? My argument is that this is not guaranteed, and it fundamentally poses a challenge for designers. Exploring how variable sensitivities map across domains (i.e. do given design parameter variations map to corresponding variations in specification values?) is a topic for future work (see Section 7.5 for further discussion). 118