MATEC Web of Conferences 139, 00016 (2017) DOI: 10.1051/matecconf/201713900016 ICMITE 2017
A new model to express and capture the design rationale in the documents
Jihong Liu1,*, Jiaji Wang1, and Kejian Wang1 1 School of Mechanical Engineering and Automation, Beihang University, 100191 Beijing, China
Abstract. As important design knowledge, design rationale（DR） knowledge plays an important role in design analysis, design reasoning and design innovation. There are generally two sources of DR knowledge, one is to capture DR in the design process, and the other is to extract DR from historical design documents, but the latter is always ignored. DR model is the foundation of DR. This paper aims at the deficiency of the previous DR model and requirements for DR knowledge acquisition from design documents. We propose design rationale knowledge hierarchy (DRKH) model. The model has three layers, design intent layer, design decision layer and design basis layer. The model also has three relationships, decomposed-into, achieved-by, refer-to. Using the DRKH model, we build the algorithm framework for extracting DR from design documents. Finally, we validate the feasibility of our model by extracting DR model from a welding robot design manual.
1 INTRODUCTION Since 1980s, many studies have made progress in the acquisition, expression and application of DR knowledge, Product design is the process of converting a person's and have made rapid progress in the DR system， purpose or need into a specific physical or tool. It is a however, there are still some problems in DR knowledge process of expressing the plan, tentative idea, and representation models. Taking into account the above problem solving method in an ideal form, through DR knowledge acquisition source and the DR specific operations. DR knowledge is used to represent knowledge expression model problems, this paper why a product or part of a product, is designed the way it proposes a new DR model for DR knowledge is [1, 2]. Therefore, it is a good way to express the representation, which aims at extracting the DR design process of product with DR knowledge. DR knowledge in design documents. Section 2 introduces knowledge includes all relevant knowledge of product previous studies on DR and text processing. Section 3 design, such as design intent decomposed for completing presents our design rationale knowledge hierarchy the product, design option made for completing the (DRKH) model. In Section 4, the design specification of design intent, discussion of design decisions in the welding robot is taken as an example, demonstrate the design process, reasons behind the design decisions, and process of extracting DR knowledge from a design so on [3]. document and expressing it with DRKH model. Section In view of the importance of DR knowledge, many 5 concludes. studies are devoted to establish DR knowledge base, i.e. express and store the acquired DR knowledge in a model and graphical language. The establishment of DR 2 RELATED WORK knowledge base consists of two processes: DR knowledge acquisition and DR knowledge representation. 2.1 DR representation model There are two sources of DR knowledge acquisition, one is stored the designer's thoughts about product design A clear and well-formed DR expression model is not during the design process; the other is extracted DR only helpful for designers to understand and use the DR knowledge from structured documents. At present, most knowledge, more conducive to the effective reuse of of the studies are committed to capture DR knowledge design knowledge. DR models are generally divided into during the design process, while ignoring the structured two categories, one is the argumentation-based DR design document which as an important source of DR representation model, and the other is the knowledge acquisition. This resulted in a great deal of intention-Driven DR representation model. At present, waste of DR knowledge. If there is an access method most DR expression models belong to that can effectively extract DR knowledge from design argumentation-based representation model. documents, it will play a significant role in promoting Argumentation-based DR representation model usually the establishment of DR knowledge base. regards the design process as a problem solving process,
* Corresponding author: [email protected] © The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0 (http://creativecommons.org/licenses/by/4.0/). MATEC Web of Conferences 139, 00016 (2017) DOI: 10.1051/matecconf/201713900016 ICMITE 2017
and considers that the design process is composed of captured document should not only have patent multiple sub processes: 1) identify design problems; 2) document, it should involve all structured design propose design alternative solutions to specific documents. Meanwhile, captured DR knowledge should problems;3) form argument that supports or opposes be graphical representation, so that designers can quickly certain design schemes;4) the decision to adopt or reject understand. Therefore, a new DR knowledge an alternative. Therefore, argumentation-based DR representation model is proposed to support the capture representation model research focuses on the process of of DR knowledge in a variety of structured design design decision generation. It records the concrete documents, and express the DR knowledge in a formation process of design decision in detail with the graphical way, such a model is beneficial to the three elements of problem, plan and argument. management and reuse of DR knowledge in subsequent The argumentation-based representation model is design. represented by IBIS (issue-based information system) which is the most mature expression in the field of DR . Many teams have developed graphical DR systems 3 Design rationale knowledge hierarchy based on the IBIS model for the DR knowledge model management. For example, graphical IBIS (gIBIS) is the first graphical IBIS tool, and Design Rationale editor 3.1 DRKH model for DR representation (DRed) is a more mature IBIS tool at the moment [4]. McCall suggested the Procedural Hierarchy of Issues (PHI) model，which simplifies the connection form in decomposed-into IBIS and makes up for the lack of dependencies between achieved-by design issues [5]. Liu et al. proposed an issue, solution refer-to and artifact layer (ISAL) representation model for DR Initial intent capture from patent documents [7]. It extends the basic sub intent sub intent Design intent layer
structure of IBIS and maps the expression of design Meta intent Meta intent thinking process to related product components, thus making up for the shortcomings of previous Design decision argumentation-based representation model. Design Design Design decision layer decision The intention-Driven DR representation model decision believes that the designer's intention will interact with
the environment and evolve with it, and hope to preserve Basis Basis
the design history by recording the evolution path of Basis Basis Design basis layer
design intention. A design intent modeling system Basis Basis developed by Arai et al. to describe the design thinking Basis process through intention, operation, and design flow. Ganesha utilize the design intent continually evaluated Fig. 1. Design rationale knowledge hierarchy model and refined to record the design process [8]. In recent The IBIS model consists of three types of elements, years, more and more researchers have realized that the namely, issues, positions, arguments, and includes the design activity is essentially an evolving cognitive relationships between the three elements. The issues are process, and the intention-Driven DR representation the discussion and reflection that needs to be done to model can reflect the cognitive characteristics of the complete a design goal, i.e., know how. Positions mean design process more accurately. the method of solving the issues, that is, know how. Arguments stand for support or opposition to positions, 2.2 Document processing that is, know why. They are connected by different rules to form an issue net. From the IBIS contains three Liu had an analysis of organized text retrieval elements can be seen IBIS model out of the product itself, technology [9]. They studied how to advance it lacks the operation of the product. The net of issues information admittance performance in document formed by IBIS only solves an issue but there is no management. Li and Ramani proposed to use shallow connection between them, and they are independent of natural language processing build an organized and each other. Although there are some problems in the semantics-based representation from design documents IBIS model, it can still provide a basis for our model. [10]. Romanowski and Nagi put forward a method to Our DR knowledge model is mainly applied to capture making generic bills of materials to backing varied DR knowledge in design documents, so we primarily design activities [11]. consider three aspects in designing model: From the literature review above, we can see that 1) The hierarchy of the model should conform to the there is still little research on capturing DR knowledge in DR structure in the design document. DR knowledge is structured design documents. Most of the existing DR generally composed of design issues, design options, system is used to capture the DR knowledge in the design decisions made by the design option, design basis design process, or for the DR knowledge retrieval, and design operations, but considering that the written of however, there are few systems for DR knowledge the design document is not record all the thoughts in the capture in design documents. We consider that the
2 MATEC Web of Conferences 139, 00016 (2017) DOI: 10.1051/matecconf/201713900016 ICMITE 2017
designer's mind, so the design options can be removed the design basis is very obvious, they usually have the from the model. words "because", "so", "therefore", "because", etc., 2) There should be a directional relationship between which express the cause. The design basis layer is an the layers of the model, and the meaning of each important level to embody the value of DR knowledge, direction is formulated. The DR knowledge captured in whether it is the basis of intent decomposition or the product design has a huge structure, and must have clear basis of choosing design decision, it is an important part direction and rules, in order to achieve efficient of the designer's thinking about the design. management and reuse of DR knowledge. 3) The model needs to be graphical. Compared to a 3.1.4 Connections between layers large number of texts, graphic language allows designers to quickly understand the content expressed by In the model in Fig.1, we define three relationships Graphical DR models. respectively as, “decomposed-into”,” achieved-by”,” According to the problems discussed above, we refer-to”. Among the relationships, propose a design rationale knowledge hierarchy (DRKH) decomposed-into
execution in the design process. Design intent is both a Arrm ssttrruuccttuurree motivation for triggering designer thinking, and ddeessiiggnn Hyyddrraauulliicc sometimes a direction for designers to think about ccyylliinnddeerr development. Design intent can be decomposed into ssttrruuccttuurree multiple levels of sub intent, when the realization of all Accttuuaattoorrddeessiiggnn sub intents, the parent's intent can achieve. The intents no need to re-decompose and can implement directly, called meta intent [12]. The design intent layer usually exists as a tree CCoonnttrroollssyysstteem sseelleeccttiioonn structure, as shown in Fig.2.The “Welding robot design” Drriivveessttrruuccttuurree Weellddiinnggrroobboott ddeessiiggnn is initial intent; “Arm structure design”, “Actuator ddeessiiggnn design”, “Control system selection”, “Wrist structure Hyyddrraauulliicc Wrriissttssttrruuccttuurree design”, “Select Transmission scheme”, “Select degrees ccyylliinnddeerrhheeaadd ddeessiiggnn of freedom and coordinate system” are sub intent; ”Arm ssccrreew sseelleeccttiioonn
structure selection”, ” Hydraulic cylinder structure”,” Hyyddrraauulliicc Drive structure design”,” Hydraulic cylinder head screw SSeelleecctt ccyylliinnddeerr TTrraannssmiissssiioonn sseelleeccttiioonn selection”,” Hydraulic cylinder selection” are meta sscchheemee intent.
SSeelleeccttddeeggrreeeessooff 3.1.2 Design decision layer ffrreeeeddoom aanndd ccoooorrddiinnaatteessyysstteem Design decision refers to the designer referring to the relevant basis, according to the evaluation criteria, put forward, analyse and compare a number of design options, and get the final solution to achieve the design Fig. 2. Design intent tree structure intent. Design decision refers to the designer referring to the relevant basis, according to the evaluation criteria, 3.2 DRKH model for DR Extraction put forward, analyze and compare a number of design options, and get the final solution to achieve the design intent. 3.2.1 Design intent extraction Our model is applied in structured design documents to 3.1.3 Design basis layer extract DR knowledge. The general form of design documentation is WORD and PDF, and structured The design basis is used to explain the reasons for design documents have header formats. Through decision making, including the basis, standards, and extensive research, we find that the title of the design tradeoffs of the designer in terms of his own expertise, document is actually the source of the design intent. But experience, preferences, and situational information. In not all titles can be used as design intent, so we will the design document, feature of the sentence which as remove the title which is can not be the design intent.
3 MATEC Web of Conferences 139, 00016 (2017) DOI: 10.1051/matecconf/201713900016 ICMITE 2017
he speci ic plan implementation rame or is sho n in ig. .
Fig. 3. The specific plan implementation framework of design intent extraction
3.2.2 Design decision extraction Fig. 5. The specific plan implementation framework of design he design decision no ledge in the design doc ment basis extraction has a characteristic，in order to reali e the design intent, the design decision is s ally a verb pl s no n phrase. n 4 Example using design document this paper, phrase str ct re based syntactic analysis is applied to e tract verb ob ect phrases rom design e se design doc ment as o r research data beca se decisions. he speci ic plan implementation framework is design doc ments are important so rce o sho n in ig. . no ledge, and very easy to get. ig. sho s the no ledge model e tracted rom the elding robot design man al sing the model. n order to veri y the applicability o the proposed model, e have done t o e periments. he e periment is to let di erent designers to e tract no ledge rom the same design doc ment ith model. e ore the designers se the model, e trained them in a ni ied ay. e analysed the model o designers, the models are e tracted rom the same design doc ment. he res lts are sho n in able .
Table 1. odel similarity data
ayer imilarity design intent layer design decision layer design basis layer
Fig. 4. The specific plan implementation framework of design rom the data e can see that the designers are decision extraction almost identical in their e traction o design intent, b t the similarity that e traction design decisions and the 3.2.3 Design basis extraction design criteria is relatively lo .
n the above analysis o the design basis, e re er to， design basis sentence s ally contain beca se...... , as a 5 Conclusion res lt o ...... , on the basis o ...... and other e planatory model is the o ndation o . A good ords and phrases. here ore, e can establish a le ical no ledge model is not only o great help to library or e pressing the reasons, and e tract the ords no ledge ac isition, moreover, it is o great rom the vocab lary library to b ild the design base signi icance or the e icient se o no ledge. layer. n this process, e dra on the related algorithms here ore, this paper aims at the de iciency o the o te t mining. he speci ic plan implementation previo s model and re irements or no ledge rame or is sho n in ig. . ac isition rom design doc ments. e propose design
4 MATEC Web of Conferences 139, 00016 (2017) DOI: 10.1051/matecconf/201713900016 ICMITE 2017
rationale no ledge hierarchy model. he model has 5. Fischer G., Lemke A., McCall R., et al., Making three layers, design intent layer, design decision layer Argumentation Serve Design, Human-Computer and design basis layer. he model also has three Interaction, 6 (3) :393-419 (1996) relationships, decomposed into, achieved by, re er to. 6. Liu Y., Liang Y., Kwong C. K., et al., A New rom the data in able , e can see that o r model Design Rationale Representation Model for needs to be improved in terms o the design decision Rationale Mining, Journal of Computing and layer and the design base layer e pression. n the Information Science in Engineering, 10(3): 1-10 ollo ing research, e ill oc s on the improvement o (2010) the model and achieve in the design doc ment to e tract 7. Ganeshan R., Garrett J., Finger S. A framework for the no ledge. representing design intent, Design Studies, 15(1): 59-84 (1994) Acknowledgements 8. Feldman, R., Text Mining: Theory and Practice, (1988) his or has been s pported by ro ect o ational 9. Liu, S., McMahon, C. A., and Culley, S. J., A cience o ndation o hina thro gh approval Review of Structured Document Retrieval (SDR) o. and ro ect o ational ey echnology Technology to Improve Information Access rogram thro gh approval o. A . Performance in Engineering Document Management, Computers in Industry, 59 (1) :3-16 References (2008) 10. Li, Z., and Ramani, K., Ontology-Based Design 1. J. Lee, K. Lai, What’s in design rationale. Information Extraction and Retrieval, Artif. Intell. Human-Computer Interaction, 6 (3) :251—280 Eng. Des. Anal. Manuf., 21 (2) :137-154 (2007) (2011) 11. Romanowski, C. J., and Nagi., A Data Mining 2. J. Lee, Design rationale systems: understanding the Approach to Forming Generic Bills of Materials in issues. IEEE Expert, 12 (3) :78-85 (1997) Support of Variant Design Activities, ASME 3. J. Lee, The 1992 Workshop on Design Rationale J.Comput. Inf. Sci. Eng., 4 (4) :316-328 (2004) Capture and Use, AI Mag., 14(2) :24-26 (1993) 12. Liu J. H., Zhan H. F., A reconstruction method of 4. J. Conklin, M. Begeman, A Hypertext Tool for the design rationale model based on design context, Exploratory Policy Discussion, ACM Transactions (2013) on Office Information Systems, 6(4): 301-331 (1988)
Weellddiinnggrroobboott
SSeelleeccttddeeggrreeeessooff SSeelleecctt Wrriissttssttrruuccttuurree CCoonnttrroollssyysstteem Arrm ssttrruuccttuurree ffrreeeeddoom aanndd TTrraannssmiissssiioonn Accttuuaattoorrddeessiiggnn ddeessiiggnn sseelleeccttiioonn ddeessiiggnn ccoooorrddiinnaatteessyysstteem sscchheemee
Hyyddrraauulliicc Hyyddrraauulliicc Hyyddrraauulliicc Drriivveessttrruuccttuurree Arrm ssttrruuccttuurree ccyylliinnddeerr ccyylliinnddeerrhheeaadd ccyylliinnddeerr ddeessiiggnn sseelleeccttiioonn sseelleeccttiioonn ssccrreew sseelleeccttiioonn ssttrruuccttuurree
EElleeccttrroomaaggnneettiicc SSeelleeccttCCaarrtteessiiaann ccoonnttrrooll TThheeddrriivvee CCoooorrddiinnaatteeSSyysstteem meecchhaanniissm ccoonnssiissttss Select Hydraulic The drive RRoobboott Select Hydraulic oofftthheehhaanndd,,wrriisstt,, The drive drive meecchhaanniissm ccoonnssiissttss drive CChhoooosseeaawrriisstt aarrm,,aannddbbooddyy oofftthheehhaanndd,,wrriisstt,, ddrriivveennssttrruuccttuurree SSeelleecctt88Mppaa aarrm,,aannddbbooddyy wiitthhoonneeddeeggrreeeessooff SSeelleecctttthheeM55 ffrreeeeddoom ssccrreewss TTeelleessccooppiicchhyyddrraauulliiccccyylliinnddeerriissuusseeddiinn hhoorriizzoonnttaalltteelleessccooppiicchhyyddrraauulliicc ccyylliinnddeerr,,VVeerrttiiccaalltteelleessccooppiicchhyyddrraauulliicc ccyylliinnddeerraaddooppttssddoouubblleeaaccttiinnggppiissttoonnccyylliinnddeerr BBeeccaauusseeooffiittsscclloossee BBeeccaauusseeiinniinndduussttrriiaallpprroodduuccttiioonn,, iinnssttiittuuttiioonnss,,moorree tthheeppuurrppoosseeooffmaanniippuullaattoorr BBeeccaauusseetthheehhyyddrraauulliicc BBeeccaauusseetthheetteennssiillee Accccoorrddiinnggttootthheerreeqquuiirreemeennttss fflleexxiibblleeaannddssoooonn ccoonnttrroolliissttooccoonnttrroolltthhee ttrraannssmiissssiioonnoouuttppuuttppooweerriiss ssttrreennggtthhooffM55iiss aannddtthheeaaccttuuaalluusseeiinniinndduussttrriiaall ccoooorrddiinnaatteeppoossiittiioonn,,aannddppaayy llaarrggee,,tthheerreessppoonnsseessppeeeeddiissffaasstt,, 880000MPPaa pprroodduuccttiioonn,,tthheeddeessiiggnnooff aatttteennttiioonnttooiittssaacccceelleerraattiioonn tthheessttrruuccttuurreeiissrreellaattiivveellyyssiimppllee weellddiinnggrroobboottuussiinnggCCaarrtteessiiaann Deeppeennddiinnggoonntthheellooaadd cchhaarraacctteerriissttiiccss aannddeeaassyyttoouussee,,ssoohhyyddrraauulliicc ccoooorrddiinnaatteerroobboo sseelleeccttiioonnpprreessssuurree,,tthhee ttrraannssmiissssiioonniissaaddoopptteedd 88Mppaameeeettsstthhee rreeqquuiirreemeennttss TThheerreeffoorree,,nnoottoonnllyyttooaacchhiieevvee tteelleessccooppiiccaarrm,,lleeffttaannddrriigghhtt rroottaattiioonn,,bbuuttaallssoottooaacchhiieevvee lliiffttiinnggmoovveemeennttss
Fig. 6. The DR knowledge model extracted from the welding robot design manual
5