Abstract
Data-driven design enables design knowledge retrieving, abstracting, reasoning and generating from massive data, and demands the integration of design theory and methodologies with machine learning algorithms. Among available data sources with potential for design, patent databases contain rich digitalized design knowledge covering all known technological domains, and have been widely exploited in engineering design research. However, one might face challenges in navigating in the vast and complex patent databases to retrieve useful design knowledge systematically and efficiently for design uses. Network analysis provides powerful tools for mapping and investigating the complex design knowledge space underlying a patent database, which could guide designers for design knowledge retrieval and analysis, as well as new design concept generation and evaluation. In this dissertation, a set of network analysis methodologies are developed to make use and sense of patent data for data-driven design. Serving as the basis of patent data-driven design, an iterative, heuristic and integrated methodology for retrieving a complete and accurate set of patents relevant to a design domain is first developed. Then, a network-based methodology is proposed to analyse the design knowledge base of the domain approximated by the collection of retrieved patents. By statistical training on historical data of a design domain in relations to the total design knowledge space, one can prescribe the potential technologies that might be adopted for next design. Furthermore, a network clustering-based methodology is proposed to classify patents in the entire patent database into home, near and far fields to a focal design problem or interest, in order to guide the search for home-, near- and far-field patent stimuli for different types of design stimulation and ideation outcomes. Additionally, a core-periphery network structure detection methodology is proposed to analyse the co-occurrence network of functions of the design precedents contained in the retrieved patents for data-driven product platform design. Each of the methodologies is demonstrated through a case study.