Journal of Engineering Design Vol. 00, No. 0, Month 2008, 1–18 CO2 DE: a decision support system for collaborative design Duck Young Kim* and Paul Xirouchakis Downloaded By: [University of Warwick] At: 18:18 11 August 2009 Ecole Polytechnique Fédérale de Lausanne (EPFL), Institut de Génie Mécanique, STI-IGM-LICP, Station 9, CH-1015, Switzerland (Received 17 December 2007; final version received 16 April 2008 ) This paper introduces an easy-to-use decision support system for COllabOrative DEsign named CO2 DE. We examine the system particularly for the design concept filtering and selection stages considering the importance of conceptual design although the proposed framework can facilitate a design task, which is recurrent in many design phases. First, the paper briefly reviews what methods can be applicable to each step of the conceptual design process, and then describes the prerequisites for mathematical methodbased design concept filtering and selection processes. Secondly, detailed explanation about the process and main modules of the proposed system are provided with a small satellite design case study focusing on (1) what the inputs and outputs of each module are, (2) what information should be specified by design participants, and (3) what results can be obtained from the system. Thirdly, the paper discusses the advantages/limitations of the system and the main difficulties and barriers for the future implementation. Finally, the paper provides a useful guideline to those who want to develop a decision support system for collaborative design by which they can minimise the need for critical refinement and serious modifications of the design at the subsequent design phase. Keywords: collaborative design; decision support system; filtering procedure; design concept selection; decision; satellite 1. Introduction Product design requires concurrent participation of many experts from multidisciplinary backgrounds because of the necessity of a large amount of different types of knowledge. For example, developing a satellite is one of the complex tasks that requires various kinds of knowledge, such as astrodynamics, thermodynamics, electronics, mechanics, space environmental engineering, manufacturing, reliability engineering, and cost engineering. In other words, mechanical and electrical designers must collaborate with environmental engineers in order to determine the overall performance of a satellite at the design stage. Moreover, there is no doubt that this collaboration should be started from the conceptual design phase because (1) design decisions are not achieved by sharing only the final results made by each design participant but rather by eliciting a satisfactory solution cooperatively under a common understanding of the design problem and (2) most of the life cycle cost of a product can be predetermined by the end of the conceptual design phase. For example, many satellite models are *Corresponding author. Email: duckyoung.kim@epfl.ch ISSN 0954-4828 print/ISSN 1466-1837 online © 2008 Taylor & Francis DOI: 10.1080/09544820802132444 http://www.informaworld.com 2 D.Y. Kim and P. Xirouchakis Downloaded By: [University of Warwick] At: 18:18 11 August 2009 not usually intended to be mass-produced, but to accomplish special mission objectives, such that the new satellite development process should be carefully planned and systematically executed from the conceptual design stage. Therefore, it is highly required to develop a decision support system for collaborative conceptual design that can help the design participants (1) find feasible design concepts with respect to various design constraints and (2) select the best design concept considering multiple design participants and their different design criteria. Let us begin with a brief review of the conceptual design process. In general, the systematic conceptual design process can be outlined by the following steps (Figure 1): • The first step of the conceptual design phase is to generate design concepts explicitly based upon: ◦ the identification of functional requirements, which can be specified by fuzzy customer needs and by means of analyses made by various company departments concerned, ◦ the establishment of function structures by decomposing the overall function into the set of sub-functions, ◦ the search for appropriate design principles (DP) for each sub-function, ◦ their combination (design concept) and the filtering of incompatible design concepts in order to elaborate the overall function on the basis of the validation of compatibilities against specified design constraints and criteria. Figure 1. Snapshot of a small satellite conceptual design process modelling. Journal of Engineering Design 3 Downloaded By: [University of Warwick] At: 18:18 11 August 2009 • During the second step of the conceptual design phase one has to select the best design concept that does minimise the need for critical refinement and serous modification at the next design phase, does best fit with the objectives of the organisation and helps the company gain competitive advantage in the long run. In short, conceptual design is a transformation process of the system specification into functional structures and engineering constraints of the overall function of a system. It involves two main steps: (1) generation of design concepts and (2) selection of the best one. There is no doubt that the major part of conceptual design relies on human creativity, namely, idea generation. Conceptual design, however, is never achieved solely by designer’s creativity but also by rational decision-making efforts in order to select the best design concept as highlighted in grey in Figure 1. Particularly, this selection procedure can be more significant when the designers try to modify/rearrange something that pre-exist(s) for improvement (adaptive design, variant design, redesign, selection design, configuration design) rather than to create something new (original design, creative design, innovative design). First, in order to generate design concepts, obtained by combining compatible sub-function DP, a systematic method to quantitatively evaluate the compatibility of a DP with the other DPs should be defined (filtering). After that, in order to select the best design concept, it is essential to adapt multiple criteria decision aiding methods (e.g. outranking methods) to the conceptual design phase (selection). In particular, they should take into account the uncertainty in the values of design parameters and criteria evaluations during the conceptual design phase. Note that CO2 DE aims to support design participants, not to replace them, during the filtering and the selection processes by means of a systematic and consistent decision-making procedure and easy-to-use system interface (e.g. Microsoft Excel™ -based user input/output interface and decision result visualisation). 2. Related work 2.1. Applicable methods Various formalised methodologies can be applicable to the five steps of the systematic conceptual design process in Figure 1: (1) product specification, (2) function structure definition by decomposition, (3) DP search, (4) combination of compatible DP by filtering, and (5) selection of the best design concept. More detailed reviews are provided for the last two steps considering the scope of this paper. Note that at this point the term ‘formalised’ implies the transformation of unstructured design activities into structured and measurable ones. (1) Product specification For the product specification step to define the overall function, the comprehensive matrix form (house of quality) of quality function deployment (QFD) methods are well founded where relationships between the customer’s requirements and technical attributes and tradeoffs among technical attributes are well described. Therefore, QFD methods can be utilised directly to convert customer requirements to functional requirements and measurable design targets during the product specification stage. (2) Function structure definition by decomposition The first concern of design verification during the conceptual design stage is to establish a valid function structure, which entails logical and physical considerations as well as matching inputs and outputs among sub-functions. In fact, the verification process of a function structure, a part of design verification, can be viewed as a functional decomposition technique, which aims to break an overall function down to functionally independent sub-functions as finely Downloaded By: [University of Warwick] At: 18:18 11 August 2009 4 D.Y. Kim and P. Xirouchakis as possible (Ullman 1992). Any flow analysis method such as bond graph and Petri nets (Bracewell and Sharpe 1996, Deng et al. 2000) is applicable to this procedure. The following three modularity methods are also applicable: function structure heuristic method (Stone et al. 2004), design structure matrix (Ulrich and Eppinger 2004), and modular function deployment (Ericsson and Erixon 1999). (3) DP search Once a valid function structure is established, DPs, to fulfil the sub-functions, have to be searched using creative and systematic thinking procedures such as brainstorming, morphological analysis, and Theory of Inventive Problem Solving (TRIZ) (Altšuller 1984). (4) Combination of compatible DPs by filtering Any combination of DPs of all sub-functions can be a potential solution for the overall function. The number of all possible combination of DPs grows drastically both with the number of sub-functions and with the number of DPs of each sub-function. Therefore, it is necessary to accelerate the search procedure in order to mitigate the verification process of the compatibility between DPs. In this case, various constraint-based approaches (e.g. Kim et al. 2006) can be usefully applied where the focus is on how to handle a large constraint network that consists of a group of variables whose values are taken from finite domains and a set of constraints on their values. Constraints here render a basis to represent compatibility between DPs. Several decomposition methods, based on the dependencies of design variables, are used to divide an original design problem into a set of small ones in order to alleviate computational complexity (Michelena and Papalambros 1997) and graph-theoretic techniques are used to search for an easier solution path (Reddy et al. 1996, Finch and Ward 1997). After a design problem is decomposed into a set of smaller ones based on variable dependencies, each small problem, formulated as a classical constraint satisfaction problem (CSP), can be practically handled by backtracking algorithms and consistency techniques (e.g. Nadel and Lin 1991). In another way of problem reduction techniques, the design constraints are divided into hard and soft ones. The former have to be satisfied under all circumstances but the latter should be taken into consideration whenever possible. In this regard, the soft constraints can be represented by user preferences. The solution procedure is achieved by relaxing the soft constraints until the solution becomes acceptable (Parunak et al. 1997). In general, the above conventional constraints-based design approaches do not deal with the characteristics of the combination problem during the filtering procedure, but rather they can be useful to search the feasible solution(s) in a given single combination of DPs (a design concept), which will be usually carried out during the subsequent embodiment and detailed design stages. In this regard, many methods for configuration design and catalogue selection design can be more attractive to the filtering procedure because of their similarity. The main goal of these methods is to select suitable components from a set of pre-defined components (components off the shelf) to achieve overall functional requirements. Mittal and Falkenhainer (1990) introduced dynamic CSP (renamed conditional CSP later) where constraints are not fixed a priori, and thus an active set of constraints and their associated variables can change dynamically in response to decisions made during the course of problem solving (e.g. requirement change such as ‘luxury car’ and ‘sport sedan’). Several researchers have agreed to this notion as a fundamental feature of configuration design problems (Snavely and Papalambros 1993, Darr and Birmingham 2000). (5) Selection of the best design concept As mentioned before, the selection process of the best design concept involves the simultaneous consideration of multiple criteria and requires concurrent participation of individuals from multidisciplinary backgrounds. Regarding the characteristics described above, a class of such problems, from the operations research and management science perspectives, Journal of Engineering Design 5 is multiple criteria decision-making (MCDM) and group decision-making (GDM), namely multiple criteria group decision-making. Naturally, many studies (e.g. Pahl and Beitz 1988, Ulrich and Eppinger 2004) on the design evaluation method have employed MCDM and GDM methods. However, it does not mean that design evaluation problems for the selection process can be directly converted to these classes of problems but rather these methods should be appropriately modified and extended considering the characteristics and assumptions of product design decision problem settings. Downloaded By: [University of Warwick] At: 18:18 11 August 2009 2.2. Summary It can be criticised that constraint-based design approaches basically presume that all constraints and variable domains are well defined, which is usually not the case in the early design stages. However, there can be little doubt that the above approaches are meaningful since the main goal of the design constraint checking procedure in the early design phase is not assigning exact values for all design variables, but rather eliminating incompatible combinations as many and as early as possible, which can be achieved even with rough variable domains and constraints. However, the main problem in applying the notion of conditional CSP directly to the filtering procedure is that the given constraints in conditional CSP (Sabin and Freuder 1996) are oversimplified, such as ‘only gasoline engine, not diesel engine, is valid for sport car design’ and ‘package option A is only compatible with battery type A’. This type of design constraints implies that the compatibility of a certain pair of DPs is given a priori. For example, it can be said that in Table 1, the following information: PL1 and ADCS2 are compatible but PL1 and ADCS1 are not, is given in advance, so that the original problem can be straightforwardly reduced. Conversely, during the filtering procedure, designers must evaluate the compatibility of DPs considering the design constraints for themselves. Furthermore, it can be said: what are changing dynamically are not the design constraints, but the domains of design variables according to what DPs are included in the combination of DPs. For example, the difference between the CSP composed by PL1, ADCS1, CDMS1, COM1, EPS1, and GS1 and the CSP composed by PL2, ADCS1, CDMS1, COM1, EPS1, and GS1 is the only domain of the design outputs (DO) in the last two columns of Table 1, but not the design constraints in Table 2. Furthermore, it is very important to note that conventional computing-based approaches (e.g. commercial optimisation software) do not fit with the selection process of the conceptual design stage because (1) each design concept is a combination of different DPs such that there is no direct relation† between different design concepts and (2) design teams should take into account multiple (often conflicting) criteria during the selection process of design concepts. In this regard, the developed system, CO2 DE aims to cope with the above limitations of current researches and computational tools for the direct usage in the filtering and selection processes of conceptual design from a practical point of view. 3. Filtering and selection problem 3.1. Prerequisites for the filtering and selection processes The results of carrying out the three steps of the conceptual design process in Section 2.1: (1) product specification, (2) function structure definition by decomposition; and (3) DP search, will † Direct relation here means derivative information for a search process. However, during the conceptual design process, an algorithm has no information about which design concept is the next to be evaluated after evaluating one design concept. 6 D.Y. Kim and P. Xirouchakis Table 1. A design catalogue for a small satellite design. Downloaded By: [University of Warwick] At: 18:18 11 August 2009 Sub-function (Fi ) Design output (DOi,k ) Design principle (DPi,j ) DP1,1 : PL1 [3 . . . 6] {0.01, 0.05, 0.1} {400, 1000} {3, 4, 5, 6} [100] [−5 . . . − 4] [−5 . . . − 4] [5 . . . 10] {2, 3, 4, 5} DP1,2 : PL2 [5 . . . 7] {0.2, 0.3} {400, 1000} {5, 10} [110] [−5 . . . − 4] [−5 . . . − 4] [8 . . . 12] {2, 3, 4, 5} point stability (deg/s) mass (g) centre of mass. X centre of mass. Y energy consumption (mWh) slot ID (location) DP2,1 :ADCS1 [0.01 . . . 0.05] [70] [−3 . . . − 2] [−3 . . . − 2] [3 . . . 5] {1, 2, 3} DP2,2 : ADCS2 [0.03 . . . 0.07] [80] [−3 . . . − 2] [−3 . . . − 2] [3 . . . 5] {1, 2, 3} DO3,1 DO3,2 DO3,3 DO3,4 DO3,5 DO3,6 data storage (Kbits) mass (g) centre of mass. X centre of mass. Y energy consumption (mWh) slot ID (location) DP3,1 : CDMS1 [630 . . . 660] [30] [−1 . . . 0] [−1 . . . 0] [4 . . . 7] {3, 4, 5} DP3,2 : CDMS2 [680 . . . 720] [40] [−1 . . . 0] [−1 . . . 0] [6 . . . 8] {3, 4, 5} Communicate with the ground stations (COM) DO4,1 DO4,2 DO4,3 DO4,4 DO4,5 DO4,6 data rate (kbps) mass (g) centre of mass. X centre of mass.Y energy consumption (mWh) slot ID (location) DP4,1 : COM1 [0.7 . . . 1.3] [70] [2 . . . 3] [2 . . . 3] [5 . . . 7] {4, 5, 6} DP4,2 : COM2 [1 . . . 1.5] [80] [2 . . . 3] [2 . . . 3] [8 . . . 10] {4, 5, 6} F5 Generate, store, regulate and distribute electrical power (EPS) DO5,1 DO5,2 DO5,3 DO5,4 DO5,5 mass (g) centre of mass. X centre of mass. Y energy consumption (mWh) slot ID (location) DP5,1 : EPS1 [250] [−1 . . . 0] [−1 . . . 0] [90 . . . 120] {1, 2, 3} DP5,2 : EPS2 [260] [−1 . . . 0] [−1 . . . 0] [110 . . . 140] {1, 2, 3} F6 Communicate with the satellite (ground station) DO6,1 number of ground station F1 Carry out the mission (payload) DO1,1 DO1,2 DO1,3 DO1,4 DO1,5 DO1,6 DO1,7 DO1,8 DO1,9 size of an image (Kbits/image) integration time, Tint(s) altitude (km) resolution (km/pixel) mass(g) centre of mass. X centre of mass. Y energy consumption (mWh) slot ID (location) F2 Determine and control the attitude (ADCS) DO2,1 DO2,2 DO2,3 DO2,4 DO2,5 DO2,6 F3 Process and distribute command (CDMS) F4 DP6,1 :GS1 {2, 3, 4} become the input information for the filtering and selection process. That is to say, this information is the prerequisite for CO2 DE and contains the following: • Function structure: the function structure of an overall function and inter-relationships (global design constraints) between the sub-functions should be identified by means of appropriate decomposition methods. • Design principles: it is necessary to search as many design alternatives for each sub-function as possible, e.g. active/passive control system for determining and controlling the attitude of a satellite. • Design variables: in contrast to the design variables which do not influence the designer’s decisions of other sub-functions (local design variables), global design variables (=local DOs) are inter-related interface variables that enable the functional interaction of the sub-functions Journal of Engineering Design Table 2. 7 Design constraints for a small satellite design. C1 (data storage): DO3,1 ≥ 452.2497 · DO1,1 /DO1,4 C2 (required no. of ground station): DO6,1 · DO4,1 · DO1,4 ≤ 1.363532 · DO1,1 C3 (pointing accuracy): sin−1 40 · DO1,4 (6378 + DO1,3 )2 − (6478)2 C4 (pointing stability): DO2,1 ≤ 0.3 sin−1 ≤ 3◦ DO1,4 (6378 + DO1,3 )2 − (6478)2 DO1,2 Downloaded By: [University of Warwick] At: 18:18 11 August 2009 C5.1 (centre of mass): DO1,5 · DO1,6 .x + DO2,2 · DO2,3 .x + DO3,2 · DO3,3 .x + DO4,2 · DO4,3 .x + DO5,1 · DO5,2 .x ≤2 DO1,5 + DO2,2 + DO3,2 + DO4,2 + DO5,1 C5.2 (centre of mass): DO1,5 · DO1,6 .y + DO2,2 · DO2,3 .y + DO3,2 · DO3,3 .y + DO4,2 · DO4,3 .y + DO5,1 · DO5,2 .y ≤2 DO1,5 + DO2,2 + DO3,2 + DO4,2 + DO5,1 C6 (mass): DO1,5 + DO2,2 + DO3,2 + DO4,2 + DO5,1 ≤ 1 kg C7 (energy consumption): DO1,7 + DO2,4 + DO3,4 + DO4,4 ≤ DO5,3 C8.1 (distance constraint for configuration): DO1,8 − DO3,5 ≤ 2 C8.2 (distance constraint for configuration): DO1,8 − DO4,5 ≤ 2 C8.3 (distance constraint for configuration): DO3,5 − DO4,5 ≤ 2 C8.4 (distance constraint for configuration): DO1,8 = DO2,5 = DO3,5 = DO4,5 = DO5,4 Note: The important local design constraints C3 is included in the list of global design constrains, without loss of generality. • • • • within a function structure. Characteristics of each sub-function can be described by a set of local DOs, e.g. size of an image (kbits) and integration time (s) for the payload subsystem of a satellite. It is worthy to note that Harmer et al. (1998) assumed in their component selection problem that the design of a product (or component) from scratch might involve satisfying a number of inter-related design requirements, but a suitable product can be selected from a catalogue by defining a few parameters (a few design variables). Design constraints: the values of local DOs are restricted by the global design constraints. Each constraint is represented as a relationship between local DOs, e.g. SUM (mass of each subsystem) < constant K(g). We will call those design constraints that restrict design variables of single sub-function local design constraints. A set of global decision-makers: a global decision-maker can be seen as an individual or a team which has a set of different global criteria reflecting his or its own interests and priorities. The group of global decision-makers consists of multidisciplinary participants or teams from engineering, manufacturing, marketing, and so on. Global design criteria: basis of decision to select the best design concept, for example, design team may want to reduce the overall energy consumption and minimise the total mass of a satellite. A set of performance evaluations of design concepts for each criterion: e.g. if a global decisionmaker has ‘total mass’as a global criterion, then it is necessary for him to define how to evaluate, 8 D.Y. Kim and P. Xirouchakis i.e. measure the level of performance for ‘total mass’ of each design concept, even if the result of measurement is too rough, non-deterministic or qualitative. Next, a preference function (a sort of mapping function) that associates each level of performance with a preference score is needed. Note here that levels of performances are very often related with the values of global design variables or output values derived by the relationships (certain equations or logical relations) among global design variables. • Two types of preference aggregations: aggregation over global criteria of each global decisionmaker and aggregation over global decision-makers. Downloaded By: [University of Warwick] At: 18:18 11 August 2009 3.2. Small satellite design problem This section introduces a small satellite design problem that will be used to show how CO2 DE can support the filtering and selection processes. According to Sarafin (1997), the payload to carry out the mission is the starting point for satellite design; it will determine satellite attitude, weight, power, communication data rates, and so on. The primary mission of the small satellite design study (Noca and Krpoun 2006, http://swisscube.epfl.ch) is the observation of the nightglow phenomena that will be measured continuously over at least one orbit. Design requirements given a priori are as follows: • Explicit ◦ Overall size: 10 × 10 × 10 cm ◦ Less than 1 kg ◦ Centre of mass must be <2 cm away from the geometric centre • Implicit ◦ Must be size-compatible with P-Pod (Picosatellite orbital deployer) ◦ Safety ◦ Environment 3.2.1. Function structure and subsystems Because of the complexity of a satellite design, this paper considers only six main sub-functions of a small satellite: (1) (2) (3) (4) (5) (6) carry out the mission (payload), determine and control the attitude (attitude determination and control subsystem), process and distribute command (command and data management subsystem), communicate with the ground stations (communication subsystem) generate, store, regulate, and distribute electrical power (electrical power subsystem) communicate with the satellite (ground station). Note that a subsystem is the embodiment of one or more sub-functions, but we assume that each subsystem has one sub-function. The combination of individual sub-functions results in a function structure representing the overall function. There can be various different function structures of a small satellite depending on the design teams’ creativity and engineering knowhow. Figure 2 illustrates a function structure example of a small satellite based on the IDEF0 method. 3.3. Design catalogue The given information for the small satellite design problem can be described by the simplified design catalogue as shown in Table 1, where each sub-function has different DPs and a set of local DOs. Downloaded By: [University of Warwick] At: 18:18 11 August 2009 Journal of Engineering Design Figure 2. 9 Function structure of a small satellite. In Table 1, each sub-function has two DPs except F6 having only one, such that 32 combinations of DPs are possible; each combination can be a potential design concept alternative for a small satellite. It is very important to note that the purpose of the filtering procedure is to select more valuable (feasible) combinations from all possible combinations in order to accelerate the search procedure of the best design concept. The main difficulty for the filtering procedure is that each combination has different domains for the DOs while the design constraints remain the same. For example, the domain of DO1,2 in the combination PL1, ADCS1, CDMS1, COM1, EPS1, and GS1 is {0.01, 0.05, 0.1} while {0.2, 0.3} in the combination PL2, ADCS1, CDMS1, COM1, EPS1, and GS1. Based on the design requirements and other design considerations (e.g. mission, budget, safety, and functional/geometric constraints), the 12 design constraints in Table 2 are considered. 4. CO2 DE 4.1. Process As illustrated in Figure 3, the overall process of CO2 DE, being essentially coherent with that of a systematic design approach in Figure 1: (1) product specification, (2) function structure definition by decomposition, (3) DP search, (4) combination of compatible DPs by filtering, and (5) selection of the best design concept, focusing on design constraint satisfaction and multiple criteria group decision-making can be outlined as follows: At the product specification and overall function decomposition phase, users (design teams) must translate customer requirements to functional requirements and measurable design targets. Downloaded By: [University of Warwick] At: 18:18 11 August 2009 10 Figure 3. D.Y. Kim and P. Xirouchakis Process of CO2 DE. Next, users, especially experts having knowledge on overall functionality define the set of subfunction and their functional structure (Figure 2) through an appropriate decomposition. Then, users search the DPs for each sub-function. At this point, it is essential to identify local DOs for each sub-function and to specify preferences on them. By matching inputs and outputs among subfunctions and transforming strict requirements (of customers and users’company) into constraints, a set of global design constraints (Table 2) should be defined. Finally, users complete constructing a design catalogue (Table 1). As mentioned in Section 3.1, the results of carrying out the above design steps become the input information to CO2 DE. At the compatible combinations filtering phase, CO2 DE extracts a set of compatible combinations of DPs (to be called k-design concepts hereafter, where ‘k’ is the user-defined number of design concept alternatives) from all possible combinations, considering global design constraints. Finally, at the selection phase of the best design concept, global decision-makers specify first global criteria. More specifically, in order to evaluate the k-design concepts with respect to the global criteria, it is needed to define how to evaluate, i.e. measure the level of performance for each Downloaded By: [University of Warwick] At: 18:18 11 August 2009 Journal of Engineering Design 11 global criterion and to define a preference function to map the level of performance to a preference score. Once each global decision-maker evaluates and determines the levels of performances of the k-design concepts with respect to all his global criteria, CO2 DE represents the preference scores for global criteria and the aggregated global preferences over the global criteria and over decision-makers. In the long run, global decision-makers are able to select the best design concept on the basis of the concept ranking given by CO2 DE. If there is no globally preferred design concept among the k-design concepts, or else some decision-making conditions for the selection process are dynamically changed in the course of the procedure (e.g. global constraints change or design teams’ preference change), it is necessary to find different design concepts, which can be achieved by an iterative procedure. Therefore, it can be said that the overall process of the filtering and selection phase of conceptual design does not exclude any interaction and iteration between the global decision-makers and local design teams for sub-functions, but instead, it just helps them to find the best design concept with respect to: global design constraints and global decision-makers’ preference scores. 4.2. Main modules of the system Figure 3 summarises what the inputs and outputs of each module are, and what information should be specified by design participants. Figure 4 illustrates the module architecture of CO2 DE. Figure 4. Module architecture of CO2 DE. 12 4.2.1. D.Y. Kim and P. Xirouchakis User interface for a design catalogue construction CO2 DE utilises Microsoft Excel™ as a tool for an input data specification because it has been widely used for efficient data handling from a practical point of view. Therefore, it can be said that input data specifications and modifications such as ‘add;’ ‘delete;’ and ‘update’ is as easy and efficient as when using Microsoft Excel™ . Once a design catalogue has been constructed and imported into the CO2 DE system, users may want to explore visually the global design constraint network to verify the relationships between sub-functions and admissible values of local DOs. In this regard, CO2 DE provides a tree-structured constraint information and a graphical user interface as shown in the workspace and the main windows in Figure 5, by which users can visualise the overall relationship between sub-functions at a glance. Downloaded By: [University of Warwick] At: 18:18 11 August 2009 4.2.2. Compatible combination filtering module As mentioned before, the filtering module helps users to extract k-design concepts from all possible combinations of DPs considering the global design constraints. Note that the domains of some design variables in Table 1 are continuous and the global design constraints in Table 2 are not only binary, so it is difficult to apply conventional binary constraint satisfaction techniques to the small satellite design problem. In this regard, first, any interval (continuous domain) will be discretised, i.e. approximated considering the given increment value. For example, the domain [3 . . . 6] for the DO1,1 of the DP ‘PL1’can be discretised to {3, 4, 5, 6} by the increment value ‘1’. The more refined the discretisation (small increment value), the more the precision increases while, on the negative side, the computational cost also increases. Secondly, it is necessary to develop a domain-specific Figure 5. Three windows of CO2 DE. Journal of Engineering Design 13 Downloaded By: [University of Warwick] At: 18:18 11 August 2009 search method to avoid full enumeration. For example, the global deign constraints in Table 2 are not so interrelated to each other, such that the overall problem can be decomposed into a set of sub-problems: (1) sub-problem with the constraints C1 , C2 , C3 , and C4 (2) with the constraints C5.1 , C5.2 , and C6 , (3) with the constraint C7 , and (4) with the constraints C8.1 , C8.2 , C8.3 , and C8.4 . For the case of binary design constraints, the reader can refer to Kim et al. (2006) about how to apply (binary) constraint satisfaction techniques to the filtering procedure. CO2 DE provides all calculation results in the ‘output window’ as shown in Figure 5. At last, CO2 DE gives a sorted list of combinations of DPs according to the number of feasible solutions, so that according to the user-defined number k of design concept alternatives, k-design concepts can be extracted from all possible combinations. For example, it can be said that the combination PL1 × ADCS1 × CDMS2 × COM1 × EPS1 × GS1, having ‘331245158400’ feasible solutions, is the most compatible combination in Figure 5. 4.2.3. Decision support module for the selection of the best design concept Initially, the tree-structured window shown at the top-left corner of the multiple criteria group decision-making (MCGDM) window in Figure 6 is intended to help users to explore information about the global decision-makers having different sets of global criteria and weights easily. The procedure of the decision support module of CO2 DE proceeds in the following sequence of Figure 6. Decision support module for the selection of the best design concept – MCGDM Window. 14 D.Y. Kim and P. Xirouchakis Downloaded By: [University of Warwick] At: 18:18 11 August 2009 numbering of Figure 6: [1] level of performance (X), [2] preference function, [3] k-design concept alternative, [4] the possible value range or discrete choices for X of the selected design concept, and [5] aggregation. [1] Level of performance (X): Users specify: • How to evaluate, i.e. measure the level of performance for each global criterion, e.g. the level of performance for the design criterion ‘energy consumption’ can be represented by the sum of each subsystem’s energy consumption. • user-acceptable value range† of X and, if necessary, target of X. Here, CO2 DE helps users to recognise the range of possible values of X under the given domains of the global design variables if X is specified by means of the global design variables. [2] Preference function: Users specify a preference function to map the level of performance to a preference score. Users can select one of the pre-defined type of preference function or specify directly a user-defined preference function, i.e. users should select one of the following option from the combo box in [2] of the MCGDM window in Figure 6, or if necessary, specify directly a preference function in the edit box in [2] after selecting the ‘User Defined’ option from the combo box: Linear(MAX), Linear (MIN), Linear(TARGET), Exponential(MAX), Exponential (MIN), Exponential(TARGET), Logarithmic(MAX), Logarithmic(MIN), and User-Defined. Users can visually verify the specified preference function through the graphical user interface at the top-right corner in the MCGDM window. [3] k-design concepts: The extracted k-design concepts from the filtering module are listed at the bottom of MCGDM window. In Figure 6, ‘3’ has been specified for the value of ‘k’. By modifying this value, users can evaluate more/less design concept alternatives. [4] The possible value range or discrete choices for X (level of performance for one criterion) of the selected design concept: Now each global decision-maker should evaluate k-design concepts with respect to his global criteria, and specify the level of performance for each global criterion of each design concept. CO2 DE helps global decision-makers to recognise the range of admissible values of X for evaluating design concepts if X is specified by means of the global design variables. Furthermore, it provides three different ways to specify values for the level of performance of a global criterion according to the global criterion’s representation type of the level of performance: • continuous (e.g. mass, energy consumption) • discrete (e.g. no. of ground stations) • single: software generated value, i.e. number of feasible solutions In Figure 6, the possible value range [19 . . . 30] for the level of performance of the design criterion ’energy consumption’is mapped to the preference score ’4.37023’of the current design concept, ’design concept 1’. [5] Aggregation: Once all global decision-makers have finished all evaluations of k-design concepts, and thus they have represented their preference scores for the k-design concepts, CO2 DE calculates the global preference scores (e.g. ‘40.41,’ ‘67.23,’ ‘41.04’ in the preference column at the bottom of Figure 7) for the k-design concepts considering weights of global decision-makers and global criteria (see top-left corner window in Figure 6). 4.2.4. Design evaluation result exploring Since it is not so easy to interpret the result of design evaluation, CO2 DE provides a spider chart, which has been widely used as a visual multiple criteria decision support tool, in order to visualise † Users can delineate a range of admissible and encouraged values for the level of performance. Downloaded By: [University of Warwick] At: 18:18 11 August 2009 Journal of Engineering Design Figure 7. 15 Design evaluation result. the evaluation result of the global decision-makers as shown in Figure 7. Finally, according to the global preference scores in Figure 7, it can be concluded that design concept 2 (combination of PL1, ADCS1, CDMS2, COM1, EPS2, and GS1) is the best design concept having the global preference score 67.23. The selected design concept will be detailed at the subsequent embodiment and detailed design stages. 4.2.5. Executive summary It is necessary to generate an executive summary based on the design evaluation results containing the detail information about the best design concept and the global decision-makers’ preferences. CO2 DE saves this information to an Excel file as shown in Figure 8. 5. Limitations and future research Some limitation of CO2 DE suggest clear direction for future research. First of all, the current version of CO2 DE can support only the last two steps of the conceptual design process: filtering and selection as shown in Figure 1. Therefore, a further study of the early phase of conceptual design concerning the generation of design concepts, as briefly reviewed in Section 2.1, should be carried out focusing on (1) how to translate product requirements into technical attributes, (2) functional decomposition methods, and (3) how to support the DP search and aid the designer’s creative idea generation. Downloaded By: [University of Warwick] At: 18:18 11 August 2009 16 Figure 8. D.Y. Kim and P. Xirouchakis Executive summary example generated by CO2 DE. The issues of how to describe the properties of each sub-function as design variables and how to represent, even create the relationships between sub-function as design constraints are prerequisites for the constraint-based approach to the filtering process. In particular, it is necessary to deal with the imprecise and incomplete information of the conceptual design stage for constraint modelling (Feng and Kusiak 1995, Lallouet and Legtchenko 2006) as modelled by different domain options for each design variable in this paper. More fundamentally, in a collaborative design environment, it can be questionable that all design participants are supposed to have the same way of representing and manipulating DPs and design constraints. Nonetheless, it can be said that the constraint-based design approach is one of the efficient ways of allowing multidisciplinary design participants to share their heterogeneous design knowledge and resolve their conflicts (e.g. Kleiner et al. 2003) and collaborative design starts with an effort to find a common way of representing and manipulating the design problem. One of the main difficulties in the development of CO2 DE is that many design constraints are non-binary, non-linear and if-then-else rules, which are difficult to handle in general (SamHaround and Faltings 1996). Considering the difficulty of developing a general purpose non-binary constraint solver, all design constraints in Table 2 have been embedded as source code in the current version of CO2 DE. For that reason, if the design constraints are changed (add/update/delete), they must be translated into source code for the new constraints and CO2 DE must be recompiled to regenerate a new version. Nonetheless, from a practical point of view, this difficulty can be diminished with a different interpretation, such as simplification of domains (e.g. ‘small’ or ‘big’ instead of detailed figures for size) and problem reduction by decomposing an overall problem into a set of smaller problems as implemented in this paper. Downloaded By: [University of Warwick] At: 18:18 11 August 2009 Journal of Engineering Design 17 A weighted aggregation method adopted for the selection process of CO2 DE. In fact, the main problem of such a method is how to aggregate the set of mono-criterion preferences to the global preference considering commensurability (measurability by the same standard or scale of values) and normalisation. In other words, the most important and difficult issue in MCDM and GDM is the allocation of the scaling constants or weights fixing the relative power of each criterion and each decision-maker in the group. Generally in the literature, it is assumed that these weights are allocated directly by a facilitator or supra decision-maker, determined informally upon consensus or agreement between decision-makers allowing interpersonal comparisons (Keeney and Raiffa 1976). However, the issue of how to systematically determine the weights should be studied carefully considering the characteristics of engineering design. First, it is necessary to specify on what grounds (e.g. certain quantitative values and their scales supported by engineering analyses and past design knowledge) design participants represent their preferences. Secondly, it should be kept in mind that the primary bases for the design participants to determine logically consistent weight values by a transformation procedure (Allen 2001, Sen 2001, See and Lewis 2004) can be experimental results of similar cases, past experience of a company, or other exclusive engineering knowhow. The allocation of weights upon agreement or consensus through negotiation or in an informal way seems to be difficult. To make the weights acceptable to all the design participants, they should be determined through transparent procedures based on tangible criteria to reflect the relevance of each design participant to the design concept selection problem. 6. Conclusions This paper has presented a framework of decision support systems for collaborative design. Since effective multidisciplinary teamwork should be started from the conceptual design phase, we have developed the system particularly for the design concept filtering and selection stages although the proposed frameword can facilitate a design task, which is recurrent in many design phases. First, design teams must define (by analysis, creative idea generation, search) function structure, DPs, design variables, design constraints, a set of global decision-makers (selected participants from the design teams), global design criteria, a set of performance evaluations of design concepts for each criterion, and preference aggregation methods. Then, based on this given information, the design system should extract compatible combinations of DPs from all possible combinations considering design constraints (filtering). Finally, these filtered combinations of DPs become the design concept alternatives from which the global decision-markers will select the best design concept with respect to the global design criteria (selection). CO2 DE has been developed based on the proposed framework to support the filtering and selection processes focusing on design constraint satisfaction and multiple criteria group decision-making. CO2 DE is currently being tested in the concurrent design facility at EPFL space centre (http://space.epfl.ch) to support a small satellite design project by students. By applying CO2 DE to the design process, it can be expected that there are the following benefits: • designers can integrate design verification (feasibility check under design constraints) and design evaluation (optimality check against design criteria) • each design team for a sub-function (e.g. satellite payload system design team or electrical power system design team) can focus on their own design problem assuming that conflicts between design teams will be identified by CO2 DE • the overall design process can be speeded up by reducing design iterations • a company can expect the structured conceptual design process (systematic reuse, consideration of different points of view, no omission of possible solutions) and a consistent decision-making during the whole design process. 18 D.Y. Kim and P. Xirouchakis Acknowledgements The authors wish to acknowledge the seed funding from the EPFL Space Center. Special thanks go to Dr. Maurice Borgeaud and Ms. Noca Muriel for introducing to us the small satellite design case study. A part of this work was presented in the proceedings of the 4th International Conference on Digital Enterprise Technology (Kim et al. 2007). This paper was carefully revised and extended to introduce the developed system, CO2 DE. Downloaded By: [University of Warwick] At: 18:18 11 August 2009 References Allen, B., 2001. On the aggregation of preferences in engineering design. Proceedings of the ASME Design Engineering Technical Conference and Compters and Informations in Engineering Conference, Pittsburgh, USA. Altšuller, G.S., 1984. Creativity as an exact science: the theory of the solution of inventive problems. New York: Gordon & Beach. Bracewell, R. and Sharpe, J., 1996. Functional descriptions used in computer support for qualitative scheme generationSCHEMEBUILDER. Artificial Intelligence for Engineering Design, Analysis, and Manufacturing, 10, 333–345. Darr, T. and Birmingham, W., 2000. Part-selection triptych: a representation problem properties and problem definition, and problem-solving method. Artificial Intelligence for Engineering Design, Analysis, and Manufacturing, 14, 39–51. Deng, Y.-M., Britton, G., and Tor, S., 2000. Constraint-based functional design verification for conceptual design. Computer-Aided Design, 32, 889–899. Ericsson, A. and Erixon, G., 1999. Controlling design variants: modular product platforms. Dearborn: Society of Manufacturing Engineers. Feng, C.-X. and Kusiak, A., 1995. Constraint-based design of parts. Computer-Aided Design, 27 (5), 343–352. Finch, W. and Ward, A., 1997. A set-based system for eliminating infeasible designs in engineering problems dominated by uncertainty. Proceedings of the 1997 ASME Design Engineering Technical Conferences, Sacramento, USA. Harmer, J., Weaver, M., and Wallace, M., 1998. Design-LED component selection. Computer-Aided Design, 30, 391–405. Keeney, R.L. and Raiffa, H., 1976. Decisions with multiple objectives: preferences and value tradeoffs. New York: John Wiley & Sons. Kim, D., Bufardi, A., and Xirouchakis, P., 2006. Compatibility measurement in collaborative conceptual design. CIRP Annals, 55 (1), 151–154. Kim, D., Bufardi, A., and Xirouchakis, P., 2007. An architecture for collaborative conceptual design support systems. Proceedings of the 4th International Conference on Digital Enterprise Technology, Bath, UK. Kleiner, S., Anderl, R., and Gräb, R., 2003. A collaborative design system for product data integration. Journal of Engineering Design, 14 (4), 421–428. Lallouet, A. and Legtchenko, A., 2006. Partially defined constraints in constraint-based design. Artificial Intelligence for Engineering Design, Analysis and Manufacturing, 20, 297–311. Michelena, N. and Papalambros, P., 1997. A hypergraph framework for optimal model-based decomposition of design problem. Computational Optimization and Applications, 8, 173–196. Mittal, S. and Falkenhainer, B., 1990. Dynamic constraint satisfaction problems. Proceedings of the 8th AAAI Conference, Boston, USA, 25–32. Nadel, B. and Lin, J., 1991. Automobile transmission design as a constraint satisfaction problem: modelling the kinematic level. Artificial Intelligence for Engineering Design, Analysis and Manufacturing, 5, 137–171. Noca, M. and Krpoun, R., 2006. SwissCube project specifications. EPFL Space Center, http://space.epfl.ch Pahl, G, and Beitz, W., 1988. Engineering design – a systematic approach. London: Springer-Verlag. Parunak, V., Ward, A., Fleisher, M., Sauter, J., and Chang, T.-C., 1997. Distributed component-centered design as agentbased distributed constraint optimization. AAAI ’97 Workshop on Constraints and Agents, Providence, Rhode Island, USA. Reddy, S., Fertig, K., and Smith, D., 1996. Constraint management methodology for conceptual design tradeoff studies. Proceedings of the 1996 ASME Design Engineering Technical Conferences, Irvine, USA. Sabin, D. and Freuder, E.C., 1996. Configuration as composite constraint satisfaction. Proceedings of the Artificial Intelligence and Manufacturing Research Planning Workshop, Albuquerque, USA, 153–161. Sam-Haround, D. and Faltings, B.V., 1996. Solving non-binary convex CSPs in continuous domains. Lecture Notes in Computer Science, 1118, 410–424. Sarafin, T.P., 1997. Spacecraft structures and mechanisms. Dordrecht: Kluwer Academic Publishers. See, T.-K. and Lewis, K., 2004. A formal approach to handling conflicts in multi-attribute group decision making. Proceedings of the ASME Design Technical Conferences, Salt Lake City, USA. Sen, P., 2001. Communicating preferences in multiple-criteria decision making: the role of the designer. Journal of Engineering Design, 12 (1), 15–24. Snavely, G. and Papalambros, P., 1993, Abstraction as a configuration design methodology. Proceedings of the 19th ASME Design Automation Conference, Albuquerque, USA. Stone, R.B., McAdams, D.A., and Kayyalethekkel. V.J., 2004. A product architecture-based conceptual DFA technique. Design studies, 25, 301–325. Ullman, D., 1992. The mechanical design process. New York: McGraw-Hill. Ulrich, K. and Eppinger, S., 2004. Product design and development. 3rd ed. New York: McGraw-Hill.