Machine learning stochastic design models

Matthews, P.C.

Abstract

Due to the fluid nature of the early stages of the design process, it is difficult to obtain deterministic product design evaluations. This is primarily due to the flexibility of the design at this stage, namely that there can be multiple interpretations of a single design concept. However, it is important for designers to understand how these design concepts are likely to fulfil the original specification, thus enabling the designer to select or bias towards solutions with favourable outcomes. One approach is to create a stochastic model of the design domain. This paper tackles the issues of using a product database to induce a Bayesian model that represents the relationships between the design parameters and characteristics. A greedy learning algorithm is presented and illustrated using a simple case study.

Citation

Matthews, P. (2005, August). Machine learning stochastic design models. Presented at 15th International Conference on Engineering Design, Melbourne, Australia

• Presentation Conference Type: Conference Paper (published)
• Conference Name: 15th International Conference on Engineering Design
• Start Date: Aug 15, 2005
• End Date: Aug 18, 2005
• Publication Date: 2005-08
• Deposit Date: Jun 3, 2008
• Publicly Available Date: Jun 3, 2008
• Series Title: Proceedings of the 15th International Conference on Engineering Design.
• Book Title: 15th International Conference on Engineering Design, ICED05, 15-18 August 2005, Melbourne, Australia ; proceedings.
• ISBN: 08582578825
• Keywords: Conceptual and preliminary design, Search and optimisation, Graphical modelling, Machine learning, Bayesian networks.
• Public URL: https://durham-repository.worktribe.com/output/1167377
• Publisher URL: http://www.designsociety.org
• Additional Information: 15-18 Aug 2005. (CD-ROM)