A Product Family Optimization Approach Using Multidimensional Data Visualization

Open Access
Author:
Slingerland, Laura A.
Graduate Program:
Mechanical Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
None
Committee Members:
  • Timothy William Simpson, Thesis Advisor
Keywords:
  • multiobjective optimization
  • product platform
  • product family
  • commonality
  • visualization
Abstract:
Product families have become an effective way for companies to provide the variety of products customers desire while keeping manufacturing costs relatively low. One of the most difficult tasks in designing a product family is deciding which components to make common and which to leave unique. There have been several approaches to this problem – some more formal than others – including the use of expert knowledge as the basis for commonality decision-making and the use of optimization algorithms. This thesis introduces an approach utilizing multidimensional data visualization to aid product family designers and create a new hybrid approach for commonality decisions. Sometimes designers struggle to recognize trade-offs because they cannot “see” inside the “black-box” models and optimization methods. Understandably, designers are often wary of solutions they cannot visualize. Using data visualization tools, designers can view the output from an optimization algorithm while also using their knowledge to “steer” the optimization in a desirable direction. The purpose of the proposed approach is to discover appropriate levels of commonality for a product family using multidimensional data visualization techniques in conjunction with optimization algorithms. The proposed approach involves five steps: (1) problem formulation, (2) optimization of all common and all unique solutions, (3) visualization of unique optimization designs with parallel coordinates, (4) optimization based on visualization results, and (5) plot of design envelope to visualize results. Multidimensional data visualization is used with “on-the-fly” design generation to enable designers to see and interact with the product family optimization process. The approach also allows designers to identify subsets of commonality rather than limit decisions to either commonality across all product variants or none at all. The end result of the approach provides designers with a plot of several platform options where trade-offs can be seen easily in the design space. Two examples are used to demonstrate this approach: (1) a family of three General Aviation Aircraft (GAA) and (2) a family of three robots. In both cases, a Pareto frontier of several platforms with varying levels of commonality is created, and although the approach is inherently subjective, it is shown to be effective and flexible; however, there are areas that can be improved. Currently, the approach has more difficulty with discrete variables compared to continuous variables due to limits of the visualization tools. Additionally, both examples used in this thesis are families of products containing three product variants. Testing should be performed on the proposed approach on families with more variants to see how well the approach scales to larger families and more design variables.