Research On Knowledge Discovery Methodology And Its Application For Product Family Development And Design

Product family development and design(PFDD)methodology is the core and fundamental of development and design for mass customization. Its objective lies in how to realize the individual customization and satisfy the individual requirements with less cost through effective product family planning, taking the product family as the management core, on the basis of the generic product family structure and regarding the product family as the organization units of the development process. The thesis aims at discovering all kinds of knowledge at different PFDD stages, furthermore applying the knowledge into engineering practices of PFDD. Such kinds of knowledge include the dominance knowledge in product family planning, similarity analysis in product family modeling, clustering knowledge and association rule discovery in product family configuration, classification and predictive knowledge discovery of configuration performance and they are discovered from different sorts of information system implemented in Enterprise. In this way, the potential of traditional PFDD can be excavated and the ability of PFDD is improved, the cost of PFDD is decreased accordingly as well.Firstly, the principles and infrastructure of Knowledge Discovery in Database for Product Family development and design (KDD-PF) is presented and the generic product family development and design process is described as well. The basic principles and formulations are addressed and the hierarchy framework is built up in which some important concepts and definitions in KDD-PF are introduced, as well as the formal representation method of the knowledge discovery process in PFDD.Secondly, along with the process of the product family development and design process, some knowledge discovery methodology and application technologies are showed as follows:(1) Combined with the traditional utility-based difference analysis for customer group, the rough analysis is carried out preliminarily. In order to carry out the more elaborate customer requirement analysis model and increase the effectiveness of decision support for product family planning, from the specific perspective of the customer agent features, the requirement information views and planning decision information tables in the mode of customer group preferences are achieved in terms of the survey questionnaires and group-decision theory, while not taking into account the conventional customer requirement features alone. Moreover, the dominant knowledge discovery for product family planning is discussed based on the extended rough set model. The methodology is verified by the market information of Kelon refrigeration and it is showed that the dominant knowledge process is more elaborate than the conventional requirement analysis, meanwhile the knowledge discovered are significant for the product family planning.(2) The distribution status and classification formulations of the parts and components of the derivation products are analyzed innovatively based on Formal Concept Analysis (FCA). In addition, BOM is the most popular and efficient representation in product family structure modeling, and most product family history information are stored as the form of BOM, therefore, the research focuses on improvement of the clustering of BOM instances and similarity matching algorithm. By applying FCA into the early stage of product family modeling, the distribution of all the components in all the product variants can be clarified and manipulated from the macro perspective, so it is more superior to the traditional Pareto method. Additionally, Similarity matching of BOM instances can increase the BOM–based product family modeling efficiency to some extent and can be utilized into the GBOM based model optimization.(3) Clustering knowledge discovery in product family configuration is discussed also, including the configuration requirement clustering and product functional specification clustering. Furthermore, the quantitative analyze model integrating variable precision rough set (VPRS) and fuzzy clustering is adopted to find out the unusual transactions in requirement configuration and to measure the performance of the mapping between requirement domain and function domain, the configuration rule discovery and configuration constraint discovery in physical domain are studied as well. The methodology is applied into the electrical power bicycles configuration and it is shown that on the one hand the discovered knowledge can enrich the traditional configuration knowledge base and increase the configuration flexibility, On the other hand quantitative analysis model provides the novel method to measure the mapping performance between requirement domain and function domain, also provides the theoretical base for discovering the unusual and unreasonable configuration.(4) The configuration performances are usually be achieved by experiment which leads to the low configuration reusability and frequent configuration changes. Based on the historical information of product family configuration, the attribute reduction algorithm based on Genetic Algorithm (GA) is proposed, as well as the integration of rough set and Artificial Neural Network (ANN). In the way the performance of the newly configuration instance can be predicted and the predicted value can be regarded as the indexes to evaluate the satisfaction degree of customer requirement. The proposed method is verified through a refrigerator product family. The results show that configuration performance prediction can turn the passive configuration into initiative configuration, decreasing the configuration period, increasing the reusability of the configuration performance prediction knowledge and the prediction precision, the reliability and reusability of the performance knowledge utilizing this method is better than the knowledge acquired by other methods.Then, the platform prototype of KDD-PF is developed including the platform selection, the main function modules and main realization interfaces.Finally, a conclusion is drawn and the trend on KDD-PF is anticipated.