Multi­Attribute Tradespace Exploration as Front End for Effective Space System Design 9 October 2009 2LT. John Richmond Greg O’Neill Jorge Cañizales Diaz MATE­CON “Multi­Attribute Tradespace Exploration with Con­Current Design” • What does it mean? • Applying a series of decision metrics (attributes) that consider the integration of all stakeholder requirements to generate a framework incorporating all qualified designs and indicating the most viable candidates. 9 Oct 2009 2 Taxonomy • MATE­CON buzzwords Decision Maker ­ Person who makes decisions that impact a system at any stage of its lifecycle Design Variable ­ Designer­controlled quantitative parameter that reflects an aspect of a concept Design Vector ­ Set of design variables that, taken together, uniquely define a design or architecture Attribute ­ Decision maker perceived metric measuring how well a defined objective is met Utility ­ Perceived value under uncertainty of an attribute Tradespace ­ Space spanned by completely enumerated design variables Pareto Frontier ­ Set of efficient allocations of resources forming a surface in metric space Exploration ­ Utility­guided search for better solutions within a tradespace Concurrent Design ­ Techniques of design that utilize information technology for real­time interaction among specialists Architecture ­ Level of segmentation for analysis that represents overall project form and function 9 Oct 2009 3 Index 1. Context of MATE­CON 2. Implementing MATE­CON 1. MATE 2. CON 3. 4. 5. 6. Alternatives Benefits Limitations Discussion 9 Oct 2009 4 Index 1. Context of MATE­CON 2. Implementing MATE­CON 1. MATE 2. CON 3. 4. 5. 6. Alternatives Benefits Limitations Discussion 9 Oct 2009 5 Context of MATE­CON • Process for Tradespace Exploration and Concept Selection (MATE). • Includes aid for Requirements Definition • Plunges forth and back into Design (CON), to win accuracy. 9 Oct 2009 Requirements Definition System Architecture Concept Generation Tradespace Exploration Concept Selection Human Factors Design Definition Multidisciplinary Optimization 6 Context of MATE­CON • Inputs: Requirements Definition • Important Stakeholders. • Set of different Concepts. System Architecture Concept Generation • Outputs: • System requirements for the Detailed Design phase. • Knowledge of the design tradespace. 9 Oct 2009 Tradespace Exploration Concept Selection Human Factors Design Definition Multidisciplinary Optimization 7 Index 1. Context of MATE­CON 2. Implementing MATE­CON 1. MATE 2. CON 3. 4. 5. 6. Alternatives Benefits Limitations Discussion 9 Oct 2009 8 Implementing MATE­CON Architecture-level Analysis True preference space 2a 1a Key decision makers User MAUT Trade space Concept generation 2b Model 3a Engineering judgment Designers 3b Analysts Verification Preference space Design-level Analysis Customer 4 Simulation (e.g. X-TOS) Solution space Firm 5 Pareto subset 6a Validation Reduced solution space 6b Sensitivity analysis True preference space User Customer 7a Proposal Firm 1b Simulation 7b Discussions Architecture­level Analysis T.S. P.S. M S.S. R.S.S. Designer MATE-CON chair 1 Fidelity feedback Subsystem chair Systems engineer 3a Real-time utility tracking Baseline 2 ICE 3b Analyst Subsystem chair Subsystem chair Images by MIT OpenCourseWare. 9 Oct 2009 Design­level Analysis 9 Index 1. Context of MATE­CON 2. Implementing MATE­CON 1. MATE 2. CON 3. 4. 5. 6. Alternatives Benefits Limitations Discussion 9 Oct 2009 10 “MATE” Overview Stakeholderi System Preference Interview Repeat for each Stakeholder Stakeholderi Single Attribute Utility Formulation Single Attribute Utility Interview Defining the System Preferences Corner Point Interview Define the System Attributes for Stakeholderi Repeat for each Stakeholder Multi­ Attribute Utility Formulation Generate Multi­Attribute Utility Functioni Tradespace n Utility Define one Set of System Attributes Create Design Vectors 9 Oct 2009 Physics­based and MAUF System Modeling Tool Metric i Image by MIT OpenCourseWare. 11 System Preference Interview & Single Attribute Utility 1. Outcome of System Preference Interview Remote Sensing Mission Attributes Communications Satellite RANK Units Range Utility Form Range Rank 4 3 1 Revisit Rate Mission Duration hour [24, 1.5] decreasing year [5,15] increasing Data Continuity (System Availability) % [30,100] increasing 9 Oct 2009 minute [240,5] decreasing 10 15 Attribute i Axis Revisit Rate (hours) 20 25 Attribute j System Availability: Single-Attribute Utility Function 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 Utility Utility (-) Utility (-) Utility 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 5 LTAN Timing [least acceptable value : most realistic, desirable value] 1 = most important attribute, 4 = least important attribute 2. Outcome of Single Attribute Utility Interview Revisit Rate: Single-Attribute Attribute Utility i Function 0 2 30 40 50 60 70 System Availability (%) Attribute j Axis80 90 100 12 Generating Single Attribute Utility Curves: The Lottery Equivalent Probability Method (LEP) � LEP Process (for one specific attribute value) Present the Interviewee with this Attributes Utility Interview Scenario Attributei (@ value) Select another Attribute Value Calculate the Utility Point Bracketing the Indifference Point Select the Probability (P*) for the Scenario Setup the Bracket Select the Preferred Situation Select another Probability Value Indifference Point (Repeat for at least 7 Attribute Values) • LEP Situation Setup Situation A Prob. = 0.5 Situation B Attributei (@ value) Prob. = P* OR Prob. = 0.5 9 Oct 2009 Attributei (@ worst value) Attributei (@ best value) 0 ≤ P* ≤ 0.5 Prob. = (1­P*) Attributei (@ worst value) 13 A Lottery Equivalent Probability Method (LEP) Scenario • Purpose: To provide context for the interviewee when selecting whether they prefer the outcomes of Situation A or Situation B in the LEP Situation Setup. • Example Interview Scenario • Attribute: resolution, Attribute Value: 4 Megapixels, Attribute Range: 1­7 Megapixels “A new optical system has been developed for a satellite that provides a higher amount of image resolution. If this optical system is used there is a chance that it could provide 7 Megapixel images versus only 4 Megapixel images when using a traditional optical system. However, the new optical system employs the use of state of the art glass manufacturing so there is a chance that the new optical system could lead to reduced image resolution (as compared to a traditional optical system). A team of engineers has studied the issue and determined that this new optical system has a P* chance of providing images with a 7 Megapixel resolution, or a (1­P*) chance of providing images with a 1 Megapixel resolution, while traditional optical systems will provide images with a 1 Megapixel resolution with a probability of 50%, and a images with a 4 Megapixel resolution with a probability of 50%. Which optical system would you prefer to use?” Situation A (Traditional Optical System) Situation B (New Optical System) 4 Megapixels Prob. = P* Prob. = 0.5 7 Megapixels OR Prob. = 0.5 9 Oct 2009 1 Megapixels Prob. = (1­P*) 1 Megapixels 14 Utility Point Calculation (from LEP Method Results) Process (for one specific attribute value) Known: The indifference point for the attribute value (i.e. P' that renders both situation A and B equally desirable to the stakeholder). Calculating the utility point for the specific attribute value is then done using Eqn. 1: 0.5 . U(Xi) + 0.5 . U(Xmin) = P'.U(Xmax ) + (1-P') . U(Xmin) The utility is calculated on a ordinal scale, where the maximum and minimum utility equal 1.0 and 0.0 respectively. Hence, Eqn. 1 becomes: 0.5 . U(Xi) + 0.5 . U(Xmin) = P'.U(Xmax ) + (1-P') . U(Xmin) 0 P' 0 U(Xi) = 2 . P' Image by MIT OpenCourseWare. 9 Oct 2009 15 Generating the Multi­Attribute Utility Function • Process (for one stakeholder) • Known: the SUAF’s for the stakeholder. • Terms: U ( X ) i ≡the i th SAUF ki ≡the i th Corner Point (SAUF Weighting Factor) K ≡the MAUF Normalization Coefficient __ U (X) ≡ the MAUF • Constructing the MAUF __ U(X ) = 1 K n (K ⋅ ki ⋅U ( X )i +1) −1+ ∏ i=1 • Capabilities of the MAUF • Determine the stakeholder aggregate utility value for a given set of single attribute utility values. • Implications • Must have the MAUF in a explicit function form • Assumptions (in addition to the 4 single attribute utility theory assumptions) • Preferential Independence: the ranking of preferences over any pair of attributes is independent of all the other attributes. • Utility Independence: The utility curve for one attribute is unique, and independent of all the other attribute utility functions. 9 Oct 2009 16 Multi­Attribute Utility Function Normalization Constant • Purpose: To ensure consistency between the MAUF and the SUAF’s. That is, ensure that the MAUF is defined over the same range as the SAUF’s (i.e. [0, 1]). • Process for Determining the MAUF Normalization Constant • Known: All the SAUF weighting factors (ki) – corner point values. • Solve Eqn. 5 for K (can be done via an iterative procedure) n K = −1 + ∏ (K ⋅ ki + 1) i =1 • Normalization Constant Ranges n if ∑k i < 1.0 then i > 1.0 ­1 < K < 0 then i = 1.0 then i =1 n if ∑k i =1 n if ∑k i =1 9 Oct 2009 K >0 K =0 17 Index 1. Context of MATE­CON 2. Implementing MATE­CON 1. MATE 2. CON 3. 4. 5. 6. Alternatives Benefits Limitations Discussion 9 Oct 2009 18 (Integrated) Concurrent Engineering Objective: to enable engineering design, tradestudies, and subsequent decisions to occur in real­ time with all design team members and critical stakeholders colocated and an emphasis placed on stakeholder feedback. Screen Screen Screen C om m Av io ni cs Pr A sN op ul ee sio de n d Reliability Conference IA&T Room and Thermal Information Stakeholder Systems Support Team Engineering Mechanical Door Team Lead Kitchen Printer Printer A/V Control Copier Mission Design Laboratory (MDL) Mission NASA Goddard Space Flight Center Flight Ops Flight Courtesy of Integrated Design Center, NASA Goddard Space Flight Center. Dynamics LVs and Software Used with permission. Administrative Attitude Cost and Technical Concurrent engineering session example: Control Support Radiation System: satellite Stakeholder: external program manager Orbital Power Printer Model Fidelity: conceptual (Phase A) Door Door Debris Session Length:’ 1 week, 5 days Daily Schedule: Design time (8 AM – noon and 1­5:30 Layout of the MDL PM); lunch (noon­1 PM) Courtesy of Mark Avnet 9 Oct 2009 Courtesy of Mark S. Avnet. Used with permission. 19 Index 1. Context of MATE­CON 2. Implementing MATE­CON 1. MATE 2. CON 3. 4. 5. 6. Alternatives Benefits Limitations Discussion 9 Oct 2009 20 Alternatives to MATE for Tradespace Exploration (TE) Tradespace Exploration Intent: enumerate candidate design concepts and ultimately select a small number of designs (called point designs), on the basis of stakeholder­ influenced criteria, to be assessed at a higher level of fidelity. Benefit­Centric TE Value­Centric TE Multiple Attribute Tradespace Exploration (MATE) Value Quantification Technique for Preference by Similarity to the Ideal (TOPSIS) Tradespace n Utility 1. MATE­CON 2. Dynamic MATE 3. System of Systems (SoS) Tradespace Exploration 4. MATE for Survivability 5. Responsive Systems Comparison (RSC) 1. Value function 2. Multi­attribute value function theory (in progress) “Traditional” TE Quantification of “Traditional” Figures of Merit (FoM) 9 Oct 2009 Metric i Image by MIT OpenCourseWare. 21 Index 1. Context of MATE­CON 2. Implementing MATE­CON 1. MATE 2. CON 3. 4. 5. 6. Alternatives Benefits Limitations Discussion 9 Oct 2009 22 Benefits • Forced design decisions’ changes, during the design phase, can be guided using the knowledge of the larger tradespace. Their impact is thus reduced. • By calculating utility gradients, counterintuitive design decisions are revealed. • Almost full automatization reduces impact of changing stakeholder expectations. 9 Oct 2009 23 Benefits • Propagating the utility metric down through the Design levels prevents pursuing a detailed design without understanding its global effects. • Proved less time and effort for a given project, and other benefits. • But the reference is to a conference paper by the author. 9 Oct 2009 24 Index 1. Context of MATE­CON 2. Implementing MATE­CON 1. MATE 2. CON 3. 4. 5. 6. Alternatives Benefits Limitations Discussion 9 Oct 2009 25 Limitations • Very different concepts are a challenge to model in 1 vector. Computer generated images of space vehicles removed due to copyright restrictions. 9 Oct 2009 26 Limitations • What do you do if the Requirements Definition tradespace is so big you cannot generate System Architecture Concept Generation but a very small Tradespace Exploration fraction of it? Concept Selection • No HF concerns are Human Factors Design Definition explicitly addressed. Multidisciplinary Optimization 9 Oct 2009 27 Limitations • Real­time design (CON) is hard to achieve for logistical and schedule reasons. • The process “doesn’t scale up”. • Not used much anymore. • Even if it’s only for early design, that needs to be done fast, the class did 12 sessions. 9 Oct 2009 28 Limitations • Doesn’t consider any “­ility”. • They all change from Concept to Concept, and even inside each one. • Their utility is usually better assessed by the engineers than by the stakeholders. • Pushing towards the frontier normally increases design cost, which isn’t considered (and can be relevant compared to manufacturing and operations cost). • Consider isoperformant frontiers. 9 Oct 2009 29 Index 1. Context of MATE­CON 2. Implementing MATE­CON 1. MATE 2. CON 3. 4. 5. 6. Alternatives Benefits Limitations Discussion 9 Oct 2009 30 Discussion Questions • Considering the architecture­level analysis and the design­level analysis that incorporate MAUT and ICEMaker, at what point do you freeze the design and move forward? • For tradespace exploration, do you think employing the metric of utility is a viable alternative to “more traditional” metrics, given the inherent advantages (e.g., aggregation of benefit) and disadvantages (e.g., ordinal nature) of utility? 9 Oct 2009 31 Discussion Questions • Stakeholders networks (utility flow) can be easily incorporated into the methodology. • What is MIT’s Generalized Information Network Analysis (GINA) method (that provided advances in modeling tradespaces)? • What is Quality Function Deployment (QFD), which is used to organize and prioritize suggested variables? • What is SMAD’s Small Satellite Cost Model? 9 Oct 2009 32 MIT OpenCourseWare http://ocw.mit.edu 16.842 Fundamentals of Systems Engineering Fall 2009 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.