Product Architecture Design and Reconfiguration using Expectation Maximization and Decision Tree Classification

Abstract

This paper addresses several of the challenges that exist in product portfolio development by incorporating data mining clustering and predictive decision tree techniques to generate the optimal product portfolio. We aim to maximize the diverse product functionality requirements of customers while concurrently minimizing product functionality overlap among products within a product portfolio. This is achieved by analyzing customer raw data and generating a robust product architecture, based on customer attribute preference information. The data mining stage utilizes a data set including asymmetric customer preference information to better model realistic scenarios. A reference product is established based on each unique product architecture that will evolve into product variants, given customer predicted preference targets. The underlying product design tradeo involves incorporating the maximum allowable flexibility in a reference product before having to generate product variants. Several influencing factors include the cost, weight or performance expectations constraining the engineering design team or the enterprise decision makers. A cell phone case study is presented that demonstrates the eectiveness of the proposed approach in generating a product portfolio comprising of reconfigurable product architectures that form the existing product families.