Research On The Methods And Key Technologies Of Data-Driven Product Conceptual Design

Enhancing innovation design ability is the important national strategy to promote innovation-driven development and build an innovative country.Innovative design is a strategic problem-solving process that focuses on the conceptual design phase,where knowledge is the source of innovation.There have been a lot of researches and achievements around knowledge-based innovative design at home and abroad,but there are many models of knowledge acquisition,organization and expression and the workload is huge,and the level of knowledge application relies more on personal experience.The conceptual design approach based on empirical knowledge and domain knowledge is difficult to access and apply cross-domain and implicit design knowledge,suffers from the limitation of design space,and does not easily produce innovative solutions.With the rapid development of big data and artificial intelligence technologies,the research of data-driven conceptual design has been promoted by fully and efficiently mining meaningful design knowledge from massive data resources.However,the data-driven conceptual design process model,how to extract and correlate cross-domain design knowledge from massive data,how to correlate explicit and implicit knowledge retrieval,and whether artificial intelligence(AI)can apply cross-domain knowledge to automatically generate design concepts need further research.To this end,the research in this paper was carried out with the general objective of a data-driven product concept design approach.This paper first outlines the current status of relevant research at home and abroad,and analyzes and summarizes the main problems that exist.Next,a data-driven conceptual design innovation principle and process model are proposed as a guide.Then a Design Knowledge Semantic Network(DKSN)is constructed to represent and relate cross-domain design knowledge.Then a graph path algorithm based on DKSN is proposed,which can retrieve both explicit and implicit knowledge associations.In addition,the Design Concept Generation Network(DCGN)is proposed to automatically generate design concepts based on design inputs.Finally,the Datadriven Conceptual Design Toolkit(DCDT)is developed.The main research of this thesis is as follows.(1)A data-driven conceptual design innovation principle and process model are proposed.A general product concept design process is analyzed.It is argued that the design knowledge embedded in highly unstructured and heterogeneous engineering text data is complex and diverse,including types of function,structure,and scientific principles,and that the inter-knowledge association relationships are highly contextualized.Taking concept design as the research object,we analyze three major innovation principles of data-driven product concept design from the perspectives of creative association theory and deep generation theory:(1)expanding design space to generate more concept solutions;(2)associating cross-domain knowledge for knowledge migration and reorganization;(3)reducing reliance on manual experience and improving design efficiency.Based on the above analysis,a data-driven conceptual design process model is constructed to provide guidance for data-driven product conceptual design.(2)A knowledge representation and association method based on the semantic network of design knowledge is proposed.A standardized design knowledge representation learning process is established to extract effective design knowledge from the large amount of unstructured and heterogeneous text data,and word embedding learning techniques are used to map complex and diverse cross-domain design knowledge to a common distributed semantic vector space.Given the highly contextual nature of knowledge associations,a new semantic association metric is proposed to quantify the strength of associations among design knowledge from the perspective of information flow and epidemic communication interactions,and a cross-domain design knowledge semantic network DKSN is constructed.The performance of the proposed DKSN is tested by comparing it with existing semantic networks.The DKSN can be used as a cross-domain knowledge base to support datadriven product concept design.(3)An explicit and implicit knowledge association retrieval method based on graph path algorithm is proposed.By analyzing the strength of explicit knowledge association in the previously constructed design knowledge semantic network DKSN,a unified knowledge association metric is established,and the strength of both explicit and implicit knowledge association is quantified.Based on the unified knowledge association metric,a unified graph path distance metric is further established,and a graph path algorithm based on DKSN is proposed.In addition,a design knowledge association intelligent retrieval framework is constructed,which can iteratively retrieve relevant explicit and implicit knowledge associations from DKSN based on design queries and expand the design space.Combined with design cases of different application scenarios,multiple knowledge association retrieval strategies are proposed to provide designers with design incentives from different perspectives to stimulate the generation of more conceptual solutions.(4)An innovative concept generation method based on design concept generation network is proposed.Based on the latest natural language generation technology Transformer,the design concept generation network DCGN is proposed by analyzing the characteristics of data-driven concept generation,i.e.,discrete design input and continuous design output.in order to support the application of product concept design,the data-driven automatic design concept generation framework is further proposed by adaptively learning the potential laws of reasoning,migration,and reorganization of cross-domain design knowledge,and can automatically generate design concepts or descriptions based on design problems or design incentives.The effectiveness and feasibility of the approach is tested and evaluated with different forms of design input examples.(5)A Data-driven Conceptual Design Toolkit(DCDT)was developed.Based on the process model and methodology of the above study,the data-driven concept design toolkit DCDT was developed,which contains four sub-tools: data collector,design knowledge organizer,design knowledge retriever,and design concept generator.Using the drone product innovation design as a case study,the application process of the tool to assist product development and design personnel in using large-scale data for product innovation design is demonstrated.