Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/6129
Title: Mining design rationale from archival documents for product design
Authors: Liang, Yan
Keywords: Product design.
Hong Kong Polytechnic University -- Dissertations
Issue Date: 2012
Publisher: The Hong Kong Polytechnic University
Abstract: Design rationale (DR) refers to an explanation of why an artifact is designed the way it is. Since DR is often regarded as crucial information to support design decisions, design analysis and design learning, DR management in engineering design is recognized as an important task. Several approaches have been developed to capture DR since the 1970s, all of which are labour intensive. However, as an increasing amount of design information is in digital format, the existing DR approaches do not have the capacity to efficiently handle DR information stored in archival design documents, which are usually in the form of unstructured or semi-structured text (e.g., design reports, service reports and test reports) that are not machine processable. Therefore, it has become an acute problem to design a computational approach to the harvesting of DR information from archival design documents and ongoing design documents for an efficient and effective storage, sharing and integration of crucial design information. This thesis proposes a novel way to approach these challenges with the primary focus on exploring text mining and machine learning techniques to discover and manage DR information from archival design documents with rich textual content. The research first proposes a new DR representation model, ISAL (issue, solution and artifact layer), which is able to support both DR discovery and DR capture. ISAL includes representing the DR of each individual document (or each design) based on issue, solution and artifact layers, and is capable of connecting multiple DRs derived from different design documents to form a DR network. This model provides a formal structure, which can be implemented by both computational approaches and human effort for DR discovering, sharing and integration. Based on the concepts of the ISAL model, this research is then to investigate and explore text mining techniques for DR discovery from digital textual design documents. Technically, five computational approaches are proposed for DR discovery. The first three graph-based approaches include a term-relation focused approach for artifact information identification, a graph-based approach with language patterns for issue summarization, and a sentence network propagation approach for solution and reason discovery. They are capable of extracting issue, solution and artifact aspects of an ISAL model by leveraging semantic and position relations between language units (e.g., words and sentences).
Graph-based approaches do not consider other language features, such as sequential data, and topic and context features. To further explore sequential modeling and topic modeling for DR discovery, a sequential language model and a hybrid topic model are proposed. The sequential language model can be applied to extract content associated with the issue aspect of an ISAL model. In this approach, each sentence is modeled as a word sequence. The modeling process is able to suggest different capability of words to indicate semantic strength associated with a specific aspect. Moreover, a hybrid topic modeling is further explored to extract salient content associated with issue aspect and solution aspect respectively. This hybrid topic model estimates the degree of a sentence associated with a specific aspect (e.g., issue aspect or solution and reason aspect) by modeling language patterns from both topic space and context space of a text collection. Experimental studies using patent documents as research data indicate that the sequential language model and the hybrid topic model proposed are capable of extracting salient sentences associated with the issue and solution and reason aspects respectively. These approaches collectively provide an effective way to extract and organize DR information into DR repositories, and they enable the DR information gathering process to be conducted in an incremental and timely manner. Furthermore, the proposed approaches can serve as a tagging and annotation process of DR information. Designers only need to confirm the DRs automatically discovered or correct them if necessary. It would improve the effectiveness of DR gathering, integration and management and make it more accessible. This research also studies a new strategy towards ISAL-based DR retrieval by utilizing the flexible DR model and its DR discovery approaches. Specifically, a DR network and DR fuzzy retrieval from multiple facets (i.e., issue, solution and artifact) are proposed and supported by exploring DR neighborhood and multi-dimension granular information between DRs. In this research, a case study using an inkjet printer design is descried to demonstrate how DR information can be represented, discovered and used based on an ISAL model. In the case study, the DR repository construction process shows how DR information can be extracted and stored using the proposed DR discovery approaches and a DR annotation process based on the ISAL model. In addition, a DR retrieval example indicates that the ISAL-based retrieval strategy is capable of integrating DR information from different views, such as key components and major technical issues. Finally, a discussion of the study as well as some suggestions for future related research is provided.
Description: xxi, 227 p. : ill. ; 30 cm.
PolyU Library Call No.: [THS] LG51 .H577P ISE 2012 Liang
URI: http://hdl.handle.net/10397/6129
Rights: All rights reserved.
Appears in Collections:Thesis

Files in This Item:
File Description SizeFormat 
b26158644_link.htmFor PolyU Users 203 BHTMLView/Open
b26158644_ir.pdfFor All Users (Non-printable)4.3 MBAdobe PDFView/Open
Show full item record

Page view(s)

291
Last Week
5
Last month
Checked on Aug 21, 2017

Download(s)

2,677
Checked on Aug 21, 2017

Google ScholarTM

Check



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.