0
Research Papers: Variability/Uncertainty in D3

Mining Process Heuristics From Designer Action Data via Hidden Markov Models

[+] Author and Article Information
Christopher McComb

Mem. ASME
Department of Mechanical Engineering,
Carnegie Mellon University,
Pittsburgh, PA 15232
e-mail: ccm@cmu.edu

Jonathan Cagan

Fellow ASME
Department of Mechanical Engineering,
Carnegie Mellon University,
Pittsburgh, PA 15232
e-mail: cagan@cmu.edu

Kenneth Kotovsky

Department of Psychology,
Carnegie Mellon University,
Pittsburgh, PA 15232
e-mail: kotovsky@cmu.edu

Contributed by the Design Theory and Methodology Committee of ASME for publication in the JOURNAL OF MECHANICAL DESIGN. Manuscript received February 20, 2017; final manuscript received June 5, 2017; published online October 2, 2017. Assoc. Editor: Harrison M. Kim.

J. Mech. Des 139(11), 111412 (Oct 02, 2017) (12 pages) Paper No: MD-17-1161; doi: 10.1115/1.4037308 History: Received February 20, 2017; Revised June 05, 2017

Configuration design problems, characterized by the assembly of components into a final desired solution, are common in engineering design. Various theoretical approaches have been offered for solving configuration type problems, but few studies have examined the approach that humans naturally use to solve such problems. This work applies data-mining techniques to quantitatively study the processes that designers use to solve configuration design problems. The guiding goal is to extract beneficial design process heuristics that are generalizable to the entire class of problems. The extraction of these human problem-solving heuristics is automated through the application of hidden Markov models to the data from two behavioral studies. Results show that designers proceed through four procedural states in solving configuration design problems, roughly transitioning from topology design to shape and parameter design. High-performing designers are distinguished by their opportunistic tuning of parameters early in the process, enabling a more effective and nuanced search for solutions.

FIGURES IN THIS ARTICLE
<>
Copyright © 2017 by ASME
Your Session has timed out. Please sign back in to continue.

References

Figures

Grahic Jump Location
Fig. 1

Example of a hidden Markov model with three states and three emission tokens

Grahic Jump Location
Fig. 2

Example truss operation sequences

Grahic Jump Location
Fig. 3

Example cooling system operation sequence with shading indicating room temperatures

Grahic Jump Location
Fig. 4

Testing log-likelihood on truss study data for models with increasing number of hidden states (error bars show ±1 S.E.)

Grahic Jump Location
Fig. 5

Transition (left) and emission (right) matrices for four-state hidden Markov model based on the data from the truss study

Grahic Jump Location
Fig. 6

Visual representation of four-state hidden Markov model based on data from truss study

Grahic Jump Location
Fig. 7

Testing log-likelihood on cooling system study data for models with increasing number of hidden states (error bars show ±1 S.E.)

Grahic Jump Location
Fig. 8

Transition (left) and emission (right) matrices for four-state hidden Markov model based on the data from the cooling system study

Grahic Jump Location
Fig. 9

Visual representation of four-state hidden Markov model based on the data from the cooling system study

Grahic Jump Location
Fig. 10

Visual representation of four-state hidden Markov model for high-performing participants from the truss study. Model for low-performing participants is structurally identical to the aggregate model (Fig. 6).

Grahic Jump Location
Fig. 11

Comparison of transmission/emission matrices for the truss study trained on: (a) high-performing data and (b) low-performing data

Grahic Jump Location
Fig. 12

Visual representation of four-state hidden Markov model for high-performing participants from the cooling systems design study. Model for low-performing participants is structurally identical to aggregate model (Fig. 9).

Grahic Jump Location
Fig. 13

Comparison of transmission/emission matrices for the cooling system study trained on: (a) high-performing data and (b) low-performing data

Tables

Errata

Discussions

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In