Recent Directions Toward Automated Life Cycle Assessment Myeon-Gyu Jeong, James R. Morrison and Hyowon Suh ISysE, KAIST 2 Presentation Overview 1. Introduction 2. LCA via CBR 3. Case study 4. Concepts for LCA at arbitrary levels of detail 5. Concluding remarks 2 1. Introduction 1. 2. 3. 4. 5. Problem definition Related work Motivation Research purpose and scope Comparison to related work 3 4 1.1 Problem definition Generic Product Development Process Mission Approval Planning Concept review Concept development System spec. review System-level design Critical design review Detail design Production approval Testing and refinement Many Iteration Cycles for Design Improvement [Design] - known [Manufacturing] - yet unknown Define part geometry Choose materials Assign tolerances Complete industrial design control doc. Piece part production processes Design tooling Define quality assurance processes Begin procurement of long-lead tooling Input Environmental Impact Evaluation Life Cycle Assessment (LCA) Preceding conditions for eco improvement Initial Design Standardized by ISO 14040~3 series Improved Design Production ramp-up 5 1.1 Problem definition Concept and General Procedure of Life Cycle Assessment Goal Definition: which products or service are assessed? How to use the result of LCA? Resources Materials Parts Ass’y Product Use Disposal Incineration Landfill Scope definition Life cycle stage Unit process Recycle & Reuse Recycle Emission to air CO2 SOx NOx Inventory analysis Measure envir. burden Global warming Impact analysis Ozone layer depletion Emission to water T-N T-P metals Acidification Considerable time and money to collect relevant data Water pollution Impact to nature and human Interpretation Review and reporting Detection of important issue Make report Check the reliability of data Critical review 6 1.1 Problem definition Limitations of LCA Collecting all relevant data and information throughout the entire life cycle at the detail design stage is impossible No matter what data is available, it requires considerable time and money In case of the product have short development cycle such as cellular phone, LCA put a burden on whole PDP [Inventory Analysis] Input 1, 2, …,n Resources output 1, 2, …,n Input 1, 2, …,n Materials output 1, 2, …,n Input 1, 2, …,n Parts output 1, 2, …,n Input 1, 2, …,n Assembly output 1, 2, …,n Input 1, 2, …,n Product output 1, 2, …,n Input 1, 2, …,n Use output 1, 2, …,n Input 1, 2, …,n Disposal output 1, 2, …,n Streamlined LCA is techniques that purposely adopt some sort of [simplified approach] to life cycle assessment 7 1.2 Related work Three basic levels of LCA (Wenzel 1998) LCA scoping Matrix LCA Life cycle stage Environmental category Scoring based on checklist [Input] [Output] Material, Energy Material (Acquisition) Air, Water, Waste Material, Energy Manufacture Air, Water, Waste Material, Energy Use Air, Water, Waste Material, Energy Disposal Air, Water, Waste Product Attribute + Existing Full LCA results Learning/Fitting Product Attributes Numerical LCA BLACK BOX (Numerical &Statistical) Compare results Matrix Operation (Weighting, Sum.) Environmental Impact Results Results (Omit indifferent process or in/output) Suggested by Graedel & Allenby(1995), Pommer(2001) Christiansen, K.(1997), Hur, T.(2003) Ines Sousa(2000), Seo, K.K(2006) No systematic procedure to select life-stage and part/module of product Hard to learning or fitting the black box and applicable to only specific product category & environmental stressor Limitations Only qualitative assessment and low accuracy of result 8 1.3 Motivation In general, most enterprises develop new products by revising or reusing the similar previous product, Power train Interior Electrics Power train Interior Electrics If we collect the LCA result of previous product, then we can estimate the LCA of new product from previous cases LCA result for regulation, certification Complicated LCA process Design output LCA result Case based reasoning for LCA Case retrieval Seat Engine Adaptation Battery Brake module 9 1.4 Research purpose and scope Generic Product Development Process Many Iteration Cycles Mission Approval Planning Concept review Concept development System spec. review System-level design Critical design review Testing and refinement Detail design Production approval Production ramp-up Design for Environment Support Case representation - FBSE expressions - Relations Case indexing - Case clustering by k-medoids CBR for LCA Case retrieval & selection - Similarity measurement and computation Case adaptation - Geometry attribute based linear modeling algorithm - Multi-regression analysis 10 1.5 Comparison to related work Standard LCA LCA scoping Matrix based Substitution of LCI DB Proposed method Precision High Medium Low Medium Medium-High Key technique Allocation, mass balance models ANN models CBR, FBSE, Regression Data type Quantitative Quantitative Qualitative/ Quantitative Quantitative Quantitative Utilization of LCI DB Yes Yes None Yes None Modeling effort required High Medium Low High Medium Required Resource Highest High Low Medium Medium Intended user Environmental expert Environmental expert Product designer Environmental expert Product designer Applicable to enterprise Low Medium High Medium High Allocation, mass Checklist based balance, quality empirical models assessment 2. LCA via CBR 1. 2. 3. 4. 5. 6. Overview of the method An FBSe representation Similarity measure Case indexing Case retrieval and selection Case adaptation 11 12 2.1 Overview of the method Similarity measure New problem w/o LCA Legend Flow Consider Function similarity Product/part specification FBS modeling Weighting Function decomposition Behavior similarity Structure (Non-numeric) Structure (Numeric) Similarity sum Find similar cluster Retrieve close case set to P Construct new cluster set Case adaptation Select Adaptation attribute Preprocessing area Case building Case formulation Old case w/ LCA Function decomposition Product/part specification LCA result FBSE modeling Clustering (k-medoid) Regression modeling Find optimal solution set Apply solution set to N Save as new case Estimated LCA result 13 2.2 An FBSE representation [New FBSE model] [Function-Behavior-State(structure) model] Solution space Problem space by Umeda Function: The purpose of the design (e.g. the purpose of a fan is to move the air) Behavior: The principle used to achieve the function (e.g. propeller fan is a kind of fan to move the air) Structure: The physical characteristics of the component (e.g. geometry size, material, color) Environmental impact: The component effect in the eleven eco-indicator 99 categories (e.g. climate change, ozone layer) [Example of FBSE expression] Two shafts have same geometry and material Fan Function To move the air Behavior Cross flow type Structure Number of blade: 20 Diameter: 50mm Length: 230mm … T1 F A Environmental impact Climate change: 0.0252Pt Radiation: 0.324mPt … T2 B To support load To transfer torque No input at “Use” stage Require lubricant at “Use” stage Different LCA result 14 2.2 An FBSE representation uctural 1 1 A (1,n) 1 A (2,n) 1 A (3,n) 1 A (m,n) uctural 3 3 A (1,n) 3 A (m,n) . effect eA s1A (m,1) s1A (m,2) Function (material) (galvanized),(sheet),(steel) fA (color) (silver) (surface),(treatment) (powder),(coating) … … s3A (m,1) s3A (m,2) s3A (m,3) (revolution),(speed) 500.0 2250.0 … … … Behavior Structure w1 … w2 … w3 w 𝑏A1 (cross) (flow) (fan) … 𝑏A2 (single) (type) … Environmental effect R1 (carcinogens) … R11 (fossil fuels) (move) (a … Function Behavior Fv … … … … Structural 2 s2A (m,1) s2A (m s2A (1,n) (mass) 22 L P 1 (raw material acquisition) … … … … s2A (2,n) (wheel), (diameter) 80 2 (part manufacturing process) (cutting),(by),(milling) 0.00525 … 0.00286 s2A (3,n) (length) 23 (assembly),(by),(rolling) 0.00422 … 0.00247 s2A (4,n) (number),(of),(blades) 26 … … … … … s2A (5,n) (number),(of),(plates) 4 6 (Disposal) … … … … s2A (m,n) … … 15 2.3 Similarity measure Attribute type Layer Modeling language Function Attribute Functional basis by Hirtz et al., at NIST Similarity Measure function Tracing the degree of kinship from hierarchical function structure Function (f) consist of pairs of words: function verb (Fv) and function object (Fo) Ex) ((move), (air)) Nonnumerical value type Behavior Attribute Standard or general engineering terminology Cosine similarity Behavior (b) consists of up to 32 phrases Ex) ((cross, flow, fan)) Point matching function Structure Attribute Product specifications or BOM terminology Structures consist of three parts S=S1×S2×S3 Numerical value type S1 is a set of two phrase vectors used for nonnumeric descriptions Ex) ((material), ((galvanized), (sheet), (steel))) S2 is a set of vectors, each consisting of a phrase and a real number Ex) ((mass), 220) S3 is a set of vectors, each consisting of a phrase and two real numbers Ex) (((revolution), (speed)), 500.0, 2250.0) Environmental effect Eco-indicator 99 method The environmental effect eE := L×P32×R11 L is the the set of life cycle stage Interval matching function 16 2.3 Similarity measure 4. Structure similarity a) Two phrase vectors used for nonnumeric descriptions 1. Function verb similarity 1 U vf c A , cB : 1 K K Bv v A K Av K Bv U c A , cB : u f i,1 , f j,1 , i 1 j 1 f v A 1 U cA , cB : o o K A KB U1 A K K u s i, 2 , s j, 2 I s i,1 , s j,1 , U1 i 1 j 1 B 1 A 1 B u f i, 2 , f j, 2 , w A B 1 1 U 2 c A , cB : 2 2 2 2 s i , 2 s j , 2 A B , I s A2 i,1 , sB2 j ,1 1 2 2 c s i ,1 c s i ,1 i 1 j 1 max A min A 2 K UA U2 A K K BU 2 2 KU B cmax sA2 i,1 : max max sc2 j, 2 I sA2 i,1 , sc2 j,1 cM 1 U cA , cB : b b K A KB 1 B b) Structural descriptions with real number values 3. Behavior similarity b 1 A p Where, the indicator function I(x, y) = 1, if x=y, and 0, otherwise. K Ao K Bo i 1 j 1 1 B 2. Function object similarity f o K UA K U B 1 1 j 1, ,32 K Ab K Bb u b i,1 , b j,1 , i 1 j 1 p A B c) The set of vectors, each consisting of a phrase and two real numbers s 3A i,3 sB3 j , 2 3 3 3 3 I s i ,1 , s j ,1 I s j , 2 , s i ,3 A B 2 B A s3 j,3 s3 i, 2 , i 1 j 1 A B 3 U 5. Overall similarity measure 3 cA , cB : K UA K U B 1 3 K UA K BU 3 3 Where, I2(x, y) = 1, if x < y, and 0, otherwise U c A , cB : wvf U vf c A , cB wof U of c A , cB U b c A , cB wU U 1 c A , cB wU U 2 c A , cB wU U 3 c A , cB . Here, wvf wof 1 and wU wU wU 1 1 2 3 1 2 3 17 2.4 Case indexing Clustering cases with representative case (𝑣𝑡 ) as the center C2 c5 c1 c3 k-medoids clustering* c4 c2 c7 c6 c5 c1 c8 𝑣2 C1 c4 c2 r2 c8 𝑣1 c6 c3 r1 c7 𝑣𝑡 ≡ arg min. dist(𝑐𝑎 , 𝑐𝑏 ) 𝑐𝑎 ∈ 𝐶𝑡 𝑐𝑏 ∈𝑐𝑡 dist 𝑐𝑎 , 𝑐𝑏 = 1 − 1 𝑛 𝑛 𝑖=1 𝑤𝑖 𝑖=1 𝑤𝑖 × sim(𝑓𝑖𝑎 , 𝑓𝑖𝑏 ) where, 𝑓𝑖𝑎 is attribute set of case a and 𝑓𝑖𝑏 is attribute set of case b 𝐶𝑡 = 𝑐1 , 𝑐2 , ⋯ , 𝑐𝑖−1 , 𝑐𝑖 , ⋯ 𝑐𝑛 dist(𝑐𝑖−1 , 𝑣𝑡 ) ≤ dist(𝑐𝑖 , 𝑣𝑡 𝑟𝑡 = max. dist(𝑐𝑖 , 𝑣𝑡 ) 𝑐𝑖 ∈ 𝐶𝑡 In out study, each indexing layer(FBS) have different data type, and some of them have nonnumeric value. In contrast to the k-means algorithm, k-medoids chooses single case as center. By introducing clustering k-medoids algorithm, we can grouping some similar cases with representative case 𝑣𝑡 as the center, and indexing the rest cases according to distance with 𝑣𝑡 18 2.5 Case retrieval and selection Case memory C2 This area is C1 r1 c6 c5 (v1) c5 𝑈 𝑣1 , 𝑁 − 𝑟 c4 c2 c3 c2 c3 (v2) C3 𝐶 c7 N c5 c7 𝑁 r3 c4 c6 c1 c1 r2 c4 𝐶𝑃𝑟 c2 c3 (v3) c6 c1 c7 r Case clustering Similarity metric: 𝑈 c𝐴 , c𝐵 , and k-medoids clustering algorithm Retrieve cluster CR 𝐶 𝑁 c7 N r c3 := {cCi: i=arg maxj U(𝑣𝑗 ,N)} Selected case set CN := {ciCR: U(ci,N) ≤ r} c6 Filtering by life cycle stage (l) and unit process (p) p 𝐶 𝑁 (l, p) ≔ c𝑖 ∈ 𝐶 𝑁 : e𝑖 l, m = p for some 𝑚 Final adaptation case set 𝐶 𝑁 (l, p) 19 2.6 Case adaptation 𝐶𝑁 c7 N r c3 For case t(j) s2t(1) (m,1) s2t(1) (m,2) For N s2N (m,1) s2N (m,2) s2t(1) (1,n) (mass) 220 s2N (1,n) (mass) 180 s2t(1) (2,n) (wheel), (diameter) 80 s2N (2,n) (length) 145 s2t(1) (3,n) (length) 230 s2N (3,n) (number),(of),(blades) 26 Where, t(j) is the original case index c6 ct(j)CN(l,p) Basis for linear regression zt(j) contains 11 real numbers for the ecological effects of that case for life cycle l and unit process p. 𝑥 2,𝑁 (m,1) 2,𝑁 𝑦𝑡(1) (m,2) 𝑦𝑁2,𝑁 (m,2) s2N (1,n) (length) 230 145 s2N (2,n) (mass) 220 180 Each row of E contains the 11 errors for the eco-impact categories for a particular case Ω10 ⋯ Ω11 2,𝑁 2,𝑁 0 𝑧𝑡 1 𝑦 𝑦 𝜀1 1 𝑡 1 ,1 ⋯ 𝑡 1 ,𝐾𝑅 1 11 Ω ⋯ Ω 1 1 ⋮ ,𝑌 = ⋮ ⋱ 𝑍= ,Ω= ,𝐸 = ⋮ . ⋮ ⋮ ⋮ ⋱ ⋮ 2,𝑁 2,𝑁 𝑧𝑡 𝑀 𝜀𝑀 1 𝑦𝑡 𝑀 ,1 ⋯ 𝑦𝑡 𝑀 ,𝐾 𝑅 Ω1𝐾𝑅 ⋯ Ω11 𝐾𝑅 𝑍 ∈ R𝑀×11 , the matrix of observed ecological effects for M cases in CN(l,p) 𝑌 ∈ R𝑀×(𝐾𝑅 +1) , KR is the number of non-null phrases Ω ∈ R(𝐾𝑅 +1)×11 is our decision variables (or independent variables) 𝐸 ∈ R𝑀×11 be our matrix of regression errors (residues) By least square error minimization, optimal decision variable values will be: Ω = 𝑌𝑇 𝑌 −1 𝑇 𝑌 𝑍 The estimated ecological effect row vector zNR1×11 for new product module N in the life cycle stage l for unit process p is: 𝑧𝑁 = 1 2,𝑁 𝑦𝑁,1 2,𝑁 ⋯ 𝑦𝑁,𝐾 Ω 𝑅 3. Case study 1. Outline of case study 2. Case memory organization by k-medoids clustering 3. Case adaptation and results 1. Case scenario 1 2. Case scenario 2 20 21 3.1 Outline of case study Target item New problem P: Cross flow fan in vehicle air purifier Plate Rolled aluminum (0.6T) Blade Rolled aluminum (0.2T) Simple Specifications Material Mass (g) 123 Wheel diameter (mm) 60 Length (mm) 230 No. of blade 26 No of plate 4 Max. RPM 2000 Impeller profile Cross flow Flow type Single Goal definition Estimate eco impact values of cross flow fan of vehicle air purifier Intended user: design engineer Scope definition Interested area is from raw material acquisition to part assembly (Upstream process) Raw material acquisition Part manufacturing Interested Area Ass’y Transportation Use Disposal 22 3.2 Case memory organization by k-medoids clustering Environmental impact was evaluated by SimaPro 7 (Commercial SW) [Case memory of fan] Classification of Impeller(blade) profile Centrifugal Axial flow Backward vaned Backward curved Forward vaned Backward inclined Strip Single Propeller Tablock (Fergas) Double Single Cross flow Mixed flow 13 2 Tubeaxial 33 3 Double 4 15 Case scenario 1 Total 100 cases were collected 10 10 5 5 Case memory C1 C2 C3 C4 C5 N Distance of P of each cluster medoid CR = C2 :={50,51,52,53,54,55,56,57,58,59,60,61} CN :={50,52,54,56,58,} Cluster No. 1 2 3 4 5 Medoid No. 97 61 41 22 68 Distance 0.8199 0.0153 0.1746 0.2601 0.6084 Case scenario 2 Cluster 2 is the closest cluster to P Case memory C1 C3 C4 C5 N CR = C3 = CN :={30,31,32,33,35,36,37,39,40,41,43,45,47,48,49} CN(2,((cutting),(by),(milling))) :={31,32,33,36,39,41,47,49} CN(2,((assembly),(by),(rolling))) :={30,37,35,40,43,45,48} 23 3.3 Case adaptation and results – Case scenario 1 0.30 0.25 Manufacturing Process - Cutting Real Estimated Manufacturing Process - Assembly 0.14 0.12 0.10 0.20 0.15 Pt Pt 0.08 0.06 0.10 0.04 0.05 0.00 0.02 0.00 Real Estimated 24 3.3 Case adaptation and results – Case scenario 1 3.5% Cutting Assembly 3.0% 2.5% Estimation Error Avg 2.56% 2.0% 1.5% 2.63% 2.21% 2.20% 2.23% 2.14% 1.99% 2.19% 2.18% 2.27% 2.32% 2.09% 1.0% 0.5% 0.0% 0.32% 0.30% 0.31% 0.35% 0.41% 0.37% 0.33% 0.33% 0.30% 0.27% 0.37% 25 3.3 Case adaptation and results – Case scenario 2 Manufacturing Process - Cutting 0.35 Real Manufacturing Process - Assembly Estimated 0.14 0.12 0.25 0.10 0.20 0.08 Pt Pt 0.30 0.15 0.06 0.10 0.04 0.05 0.02 0.00 0.00 Real Estimated 26 3.3 Case adaptation and results – Case scenario 2 8.0% Cutting Assembly 7.0% Avg 6.88% Estimation error 6.0% 5.0% 4.0% 6.61% 4.85% 6.20% 5.97% 5.57% 5.36% 6.03% 6.00% 6.25% 6.39% 5.77% 3.0% 2.0% 1.0% 0.0% 0.85% 1.15% 0.84% 1.14% 1.22% 1.13% 0.89% 0.91% 0.82% 0.76% 1.02% 4. Concepts for LCA at arbitrary levels of detail 27 28 4. 4. Concepts for LCA at arbitrary levels of detail Product A FA 𝐹𝐴1 ⊇ 𝐹𝐴3 , 𝐹𝐴4 𝐹𝐴1 = {𝐹𝐴3 , 𝐹𝐴4 } Sub-product A1 FA1 Sub-product A2 FA2 The function of sub-product(FA1) is comprised of functions of component(FA3, FA4) For example, {(move), (air)} and {(rotate), (fan)} is subset of {(make), (wind)}. Only leaf nodes of hierarchy have behavior, structure and environmental effect description, Component A3 Component A4 Component A5 Component A6 FA3, BA3, SA3 FA4, BA4, SA4 FA5, BA5, SA5 FA6, BA6, SA6 because leaf nodes indicate single component has detailed behavioral and structural attributes. And FBSE attribute is subordinate to upper node. Electric fan {(make), (wind)} Therefore, we can describe BA1 with BA3 and BA4. However, if the function of new problem N is not fully decomposed, we can not search the environmental effect from lower level, because there are no B, S information. Under these situation, construction the generalized functional hierarchy for specific product family will gives some help to find anticipated sub functions and associated environmental effect. In other words, with these generalized functional hierarchy, we can anticipate and support the design process with only high level functional attribute. The figures in next pages show these concepts. Fan {(move), (air)} Motor {(rotate), (fan)} 29 4. 4. Concepts for LCA at arbitrary levels of detail Old case A Sub-product A1 FA1 Old case B Product B FB Product A FA Sub-product B1 Sub-product A2 FA2 FB1, BB1, SB1, EB1 Sub-product B2 FB2 Component A3 Component A4 Component A5 Component A6 Component B3 Component B4 FA3, BA3, SA3, EA3 FA4, BA4, SA4, EA4 FA5, BA5, SA5, EA5 FA6, BA6, SA6, EA6 FB3, BB3, SB3, EB3 FB4, BB4, SB4, EB4 B1 is instance of C6 A4 is instance of C4 Model C is the generalized functional hierarchy for specific product family Model C FC Case memory Sub-model C1 FA1 Sub-model C2 FA2 A3 is instance of C3 Sub-model C3 Sub-model C4 FA3, BA3, SA3, EA3 FA4, BA4, SA4, EA4 cases cases cases cases cases cases Sub-model C5 FA5 Sub-model C7 Sub-model C8 FA5, BA5, SA5, EA5 FA6, BA6, SA6, EA6 cases cases cases cases cases cases Sub-model C6 FA6, BA6, SA6, EA6 cases cases cases B4 is instance of C8 30 4. 4. Concepts for LCA at arbitrary levels of detail If the function of product N is not fully decomposed, we cannot estimate the E of N1, N2 and N5. Model C FC Product N FN Sub-model C1 FA1 Sub-model C2 FA2 Sub-product N1 FN1 Sub-model C3 Sub-model C4 FA3, BA3, SA3, EA3 FA4, BA4, SA4, EA4 cases cases cases cases cases cases Sub-model C5 FA5 Sub-model C7 Sub-model C8 FA5, BA5, SA5, EA5 FA6, BA6, SA6, EA6 cases cases cases Sub-product N2 FN2 Sub-model C6 FA6, BA6, SA6, EA6 Component C3 Component C4 Sub-product N5 Component c6 FN3, BN3, SN3, EN3 FN4, BN4, SN4, EN4 FN5 FN6, BN6, SN6, EN6 cases cases cases cases cases cases However, if the product N is subset of Model C, C3, C4, C6, C7 and C8 will be anticipated lower function. After confirm the sub functions and associated behavior, structure, finally we can estimate environmental effect with LCA via CBR process. Component C7 Component C8 FN5, BN5, SN5, EN5 FN6, BN6, SN6, EN6 5. Concluding remarks 31 32 Concluding remarks 1. Introduction 2. LCA via CBR 3. Case study 4. Concepts for LCA at arbitrary levels of detail 32 Appendix 33 34 * Functional basis reconciled function set Correspondents Tertiary Secondary Isolate, sever, disjoin Detach, isolate, release, sort, split, disconnect, subtract (1,1,1) Divide Refine, filter, purity, percolate, strain, clear (1,1,2) Extract Cut, drill, lathe, polish, sand Primary Primary Secondary Separate Branch (1,0,0) Control Magnitude Actuate (1,1,0) (6,0,0) Regulate (6,14,0) (1,2,0) (2,3,0) Import Channel Change Dispose, eject, emit, empty, remove, destroy, eliminate (2,4,0) Export (2,0,0) (6,15,0) Carry, deliver Transfer Control, equalize, limit, maintain (6,14,11) Increase Allow, open (2,5,4) Transport Conduct, convey (2,5,5) Transmit Guide Move, relocate (2,6,0) Translate (2,6,7) Increment (6,15,13) Amplify, enhance, magnify, multiply Condition (0,0,0) Rotate Assemble, fasten (3,7,9) Join Attach (3,7,10) Link Couple connect (3,7,0) (3,0,0) (6,16,17) Prevent Provision Store (7,0,0) (7,17,0) Contain (7,18,0) Collect (6,16,18) Steady (4,9,0) Stabilize Constrain, hold, place, fix (4,10,0) Align, locate, orient (4,11,0) Position Condense, create, decode, differentiate, digitize, encode, evaporate, generate, integrate, liquefy, process, solidify, (5,12,0) transform Secure Convert Signal Sense (8,0,0) (8,19,0) Support (7,17,20) Shield, insulate, protect, resist Capture, enclose Absorb, consume, fill, reserve Feel, determine (8,19,21) Detect Indicate (8,19,22) Discern, perceive, recognize Identify, locate Announce, show, denote, record, register (8,20,0) Track Convert (8,21,0) Display (5,0,0) Process * Reference: NIST Technical Note 1447 “A Functional Basis for Engineering Design: Reconciling and Evolving Previous Efforts” Disable, turn-off Provide, replenish, retrieve Measure (4,0,0) Prepare, adapt, treat Accumulate (7,17,19) Supply (3,8,0) Mix Attenuate, dampen, reduce Compact, compress, crush, pierce, deform ,form End, halt, pause, interrupt, restrain Inhibit Associate, connect (6,15,16) Stop (6,16,0) Allow DOF (6,15,14) (6,15,15) Shape Functional basis Direct, shift, steer, straighten, switch (2,6,6) Close, delay, interrupt Adjust, modulate, clear, demodulate, invert, normalize, rectify, reset, scale, vary, modify Decrement (2,5,0) Advance, lift, move Add, blend, coalesce, combine, pack Enable, initiate, start, turn-on Distribute Form entrance, allow, input, capture (2,6,8) (6,13,0) Decrease (6,14,12) Diffuse, dispel, disperse, dissipate, diverge, scatter Constrain, unfasten, unlock Correspondents Remove (1,1,3) Spin, turn Tertiary (8,20,23) (8,20,24) Mark, time Emit, expose, select Compare, calculate, check 35 *k-medoids clustering algorithm The k-medoids algorithm is a clustering algorithm related to the k-means algorithm and the medoid shift algorithm. In contrast to the k-means algorithm, k-medoids chooses data points as centers (medoids or exemplars). 1. Arbitrary select k of the n data points as the medoids 2. Associate each data point to the closest medoid, and calculate total cost of each cluster 3. Swapping medoid and random case, and calculate total cost 4. Finalized cluster set In our research, similarity measurement can be defined as the sum of functional distance, behavioral distance and structural distance. However each indexing layer have different data type, and some of them have nonnumeric value. Therefore, k-medoids clustering algorithm is appropriate to us than k-means.