An object-oriented modeling framework for representing uncertainty in early variant design Rose Crossland, Jon H. Sims Williams, Chris A. McMahon Abstract This article presents a framework for the repre- sentation of uncertainty in the early design of complex adaptive products such as automobiles. The core of the framework is an object-oriented approach in which design objects and their inter-relationships may be modeled, and in which both the design attributes and the product structure may be uncertain. Relationship objects allow product vari- ants and design alternatives to be represented. In addition to the design model, derivation methods for design attributes may be modeled, and methods may be incorporated to allow the deterministic or probabilistic computation of attributes. The modeling framework is the basis of a risk modeling tool, RiTo, in which Monte Carlo simulation is used to compute estimates for costs and other design attributes together with their probability of achievement in the final design. Uncertainties may be aggregated and levels of uncertainty in different parts of the model may be continually analysed and assessed. The framework also provides a mechanism for accumulating product knowledge, in particular knowledge concerning relationships between elements of part and assembly models, product volumes and manufacturing considerations. Keywords Risk in design, Uncertainty in design, Object modeling, Monte Carlo simulation, Variant design 1 Introduction In the early stages of a team-based design project, it is difficult to assess the emerging costs and performance parameters of the new design. Knowledge is distributed between team members, and the design decisions are subject to many types of uncertainty (Ullman and DÕAm- brosio 1995; Chen and Yuan 1997). Uncertainty—and hence risk—arises from many sources including design decisions as yet unmade, uncertain environmental factors, lack of specification detail, missing information, the use of rules-of-thumb to estimate design attributes and the existence of alternative behaviour models. Thurston and others (Thurston 1991; Thurston and Tiefu 1991; Ullman and Herling 1995; Carnahan et al. 1994) proposed decision support techniques for design evaluation under uncertainty. However, few tools currently exist to support sharing of complex and uncertain early design information, and thus many critical early design decisions are made on the basis of a sub-set of the avail- able information. Bilgic ¸ and Rock (1997) observe that the lack of support for resource, performance and risk man- agement is a significant limitation of the current genera- tion of product data management (PDM) systems. This paper describes an application of a risk modeling frame- work that allows designers to model a product in order that risks, both cost and technical, may be assessed during this early design phase. Product modeling systems have the aim of allowing the development of integrated, nor- mally computer-based, models to support all of a product life cycle (McMahon and Browne 1998). The framework is intended to be a test bed for exploring ideas for incorpo- ration of the representation of uncertainty into such sys- tems. It is closely related to ideas from recent work in product structuring (Baldwin and Chung 1995; McKay et al. 1996; Duffy et al. 1996; McKay 1997; Tiihonen et al. 1998; Rosenman and Wang 1999), but by combining rep- resentation of product structure with uncertain product attributes, attribute derivation relationships and a general simulation capability it allows a probabilistic approach to engineering design modeling under uncertainty (Cross- land 1997; Crossland et al. 1995). This approach, like Wood and AntonssonÕs fuzzy-based ‘‘method of impreci- sion’’ (Antonsson and Otto 1995; Wood et al. 1989), facilitates set-based concurrent design (Ward et al. 1995) and also aims to support rapid early model building through re-use of classes of design objects. The framework employs an object-oriented approach that allows a design team to build a shared model of a generic structure for a particular product type, of deriva- tion methods for design attributes for these products and of a new design as it emerges, all incorporating uncer- tainty. The transparent incorporation of probabilistic uncertainty into an object-oriented product modeling framework is believed to be novel. This approach makes it Research in Engineering Design 14 (2003) 173–183 DOI 10.1007/s00163-003-0039-z 173 Received: 1 November 2001 / Revised: 11 September 2002 Accepted: 20 May 2003 / Published online: 14 October 2003 Ó Springer-Verlag 2003 R. Crossland, J. H. S. Williams Department of Engineering Mathematics, University of Bristol, Bristol, BS8 1TR, UK C. A. McMahon (&) Department of Mechanical Engineering, University of Bath, Bath, BA2 7AY, UK E-mail: enscam@bath.ac.uk Fax: +44-1225-386928 This work was carried out as part of the STARTED project (grant reference GR/J53867), which was supported by the Department of Trade and Industry/Engineering and Physical Sciences Research Council CSCW program. Thanks are also due to our industrial partners Cegelec Projects Ltd, Rover Group Ltd and Ove Arup and Partners. Original paper