Defining product specifications to meet customers’ preferences is a crucial and challenging task for custom product design. An efficient specification defining method should take both product structures and customers’ preferences into consideration. Because customers’ preferences depend largely on factors such as product attributes and external parameters, conventional specification definition methods in deterministic form fall short of providing adequate approaches to represent and manipulate the probabilistic nature of customers’ preferences. They often suffer from low efficiency, lack of intelligence to adapt to different customers’ inputs, being unable to provide guidance to users who have little domain knowledge, etc. These technical issues have hindered the development of custom product design and mass customization. To solve these issues, Bayesian network is deployed to represent the product physical structure and the likelihood of the customers’ potential preferences among components. The specification defining is modeled as an uncertainty elimination process and an information theory based algorithm is applied to obtain customers’ target product configuration. The idea is to sequentially select the most relevant component for a customer to specify from the remaining components pool based on his previous step’s specification. An inference module is also proposed which can recommend possible product configurations to help customers find what they want quickly. Thus a customized 1-to-1 specification defining procedure is provided and the final configuration can converge to a customer’s target with fewer interactions between the customer and product design team. This paper explores the framework of product specification defining in uncertain domain and offers a new angle to advance design for mass customization (DfMC), and possibly the Design for Manufacturing in general.
Configuration for mass customization: how to extend product configuration towards requirements and process configuration
