Dr Peter Matthews
(January - March 2008)
Dr Peter Matthews holds degrees in Mathematics (1994), Computer Science (1995) and Engineering Design (2002), which were all taken at Cambridge University. He stayed on at Cambridge to continue with postdoctoral research work at the Engineering Design Centre, continuing his research in analysis and synthesis methods for the conceptual engineering design process. During this period, he undertook brief secondments with BAE Systems and Rolls-Royce (Aerospace) to further his understanding of the industrial challenges and requirements of engineering design. He was subsequently appointed to an engineering lectureship at Durham University in 2003.
Dr Matthews has published on the challenges of creating and exploiting useful quantitative models of the early engineering design process. The early, conceptual, design process is characterised by its fluid nature and thereby its resistance to traditional engineering models which require relatively concrete data to be of any use. His earlier work (2002) used Self Organising Maps and non-hierarchical clustering algorithms as a means for guiding designers to extract engineering heuristics for conceptual design from databases of prior designs. This was followed by an approach to provide richer heuristics by using a Genetic Programming approach (2003). The core algorithm for this data analysis process was filed as British and European patents.
Since Dr Matthews' appointment to Durham, he has subsequently refined his representation methodology towards a stochastic approach (2006). This work was funded by a Nuffield Foundation Newly Appointed Lecturers award in collaboration with Rolls-Royce. The research has identified and developed a significantly more efficient algorithm for automatically learning the causal stochastic domain rules, thereby being able to rapidly construct Bayesian Belief Networks. In addition, the work has also developed a prototype interface to allow an engineering designer to exploit these rules and models when developing a new product.
During his IAS Fast Track fellowship, Dr Matthews will be considering the interdisciplinary aspects of stochastic models, specifically from an agent-based perspective. By investigating alternative disciplinary approaches to agent-based and stochastic decision modelling, this research will seek to provide a more robust foundation for modelling human behaviour within engineering design. This, in turn, will provide a foundation for creating better decision support tools in ambiguous domains. As part of the work at the IAS, Dr Matthews will be collaborating with Prof Coolen (Mathematical Sciences), Prof Vitanov (School of Engineering) and Prof Wright (Durham Business School) and will be publishing on the state of the art in stochastic modelling methodologies across different disciplines and how these ideas can define future stochastic decision support methodologies.
As part of this work, regional and national industries will be invited to an IAS sponsored seminar to discuss the potential benefits and challenges that are presented by stochastic decision making methodologies. As part of its agenda, it will aim to explore realistic methods and model representations for supporting human decision makers.
IAS Fellow's Public Lecture - Modelling the Engineering Design Decision Process
Models of the engineering design process assume that the human designer is a rule-based agent and primarily focus on how they access and use technical knowledge for design. They are not able to represent the subjectivity bought to the design process by human agents. This subjectivity is particularly important during the earlier, conceptual, stages of the design process. Agenda-setting research in this field is investigating the potential of stochastic approaches for agent-based models to address this challenge.
However, these approaches are still experimental and little work has been done to learn how other disciplines have approached similar problems. By investigating alternative disciplinary approaches to agent-based and stochastic decision modelling, this research will seek to provide a more robust foundation for modelling human behaviour within engineering design.