Computer-Aided Creative Mechanism Design

Computer-aided Creative Mechanism Design

Boi Faltings and Kun Sun Laboratoire d'Intelligence Artificielle (DI) Swiss Federal Institute of Technology (EPFL) IN-Ecu blens

1015 Lausanne, Switzerland

Tel. +41-21-693-2738, FAX: +41-21-693-5225

E-Mail: faltings@lia.di.epfl.ch

Knowledge-based Design Knowledge­ based design systems, such as ICAD, are cur­ rently used in the context of large engineer­ ing projects. They provide design knowl­ edge in the form of prototypes which can be instantiated and updated automatically ac­ cording to the design context. Many design­ ers are sceptical of the technique because the prototype library creates a preference for es­ tablished solutions and thus inhibits innova­ tion.

A popular saying claims that "innovation is 1 % inspiration and 99 % perspiration." While it is usually very difficult to endow a computer with sufficient breadth of knowl­ edge to provide the inspiration, tools which support innovation by automating the per­ spiration seem much more feasible. The goal of our research are techniques which provide such support using artificial intel­ lligence techniques.

This video presents the FAMING system, a program which illustrates an approach to supporting creative design on the example domain of elementary mechanisms. This class includes devices such as gears, escape­ ments or ratchets which consist of two parts and achieve their function by interaction of their shapes. They are very easy to model and simulate, but variations of part shapes give a very rich variety of designs. In fact, novel devices are invented to this day. El­ ementary mechanisms are thus an ideal do­ main for studying how creative design could be supported by computer, even if the do­ main is too specialized for commercially vi­ able design tools.

Innovation in knowledge-based design As innovations are by definition unknown to the designer of a computer tool, knowl­ edge to support them cannot be very spe­ cific, but must be based on general prin­ ciples which remain valid even in innova­ tive solutions. Approaches using model­ based reasoning ([Williams, 1990; Neville and Weld, 1993]) work well whenever inter­ actions between the individual models can be precisely identified (the no-function-in­ structure principle).

Kinematic pairs achieve their function through the possible contacts between ele­ ments of the part shapes. Because all con­ tacts have to be integrated into a common geometry, they interact in very unstruc­ tured ways: any contact can subsume (in­ hibit) any other one and make it impossible. Kinematics therefore does not respect the no-function-in-structure principle. Thus, while Joskowicz and Addanki ([Joskowicz and Addanki, 1988]) have attempted to ap­ ply methods of model-based design to kine­ matic pairs, their method is restricted to simple cam-like mechanisms and incapable of producing devices such as escapements or ratchets.

Humans often produce creative designs by reinterpreting known solutions in a novel way [Wills and Kolodner, 1994]. FAMING supports this form of creativity by two major innovations over existing techniques. The first is the use of case-based reason­ ing as a main source of design knowledge to handle the context-dependence inherent in geometry; FAMING is one of the first programs which actually generate novel devices by combining and adapting previous cases. The second is that FAMING uses a general theory of qualitative kinematics ([Faltings, 1990; 1992]), capable of reasoning about any imaginable qualitative function, to provide the knowledge for case adaptation and com­ bination.

As input to FAMING, the designer provides a formal specification of the desired func­ tions, and examples of shapes that can im­ plement aspects of them. In agreement with common design methodology (e.g., [Pahl and Beitz, 1984]), functions are qualitative and indicate an interpretation of the exam­ ple device. They are specified as logical ex­ pressions with reference to an envisionment of all possible behaviors of the device they interpret. In contrast to fixed functional vo­ cabularies used in current intelligent CAD systems, this qualitative formalism can ex­ press any imaginable qualitative function in­ cluding creative ideas unknown to the pro­ grammer.

The theory of qualitative kinematics is used to map specifications into constraints on the part shapes provided in the exam­ ples, and thus their adaptation into novel uses. The new prototypes are then added to the knowledge-based system's vocabu­ lary ([Faltings and Sun, 1993; Sun and Falt­ ings, 1994]) and can subsequently be in­ tegrated in designs using all functionali­ ties provided by the intelligent CAD tool. FAMING thus makes it possible to integrate creative designs in a knowledge-based CAD system.

FAMING Results FAMING has successfully designed several devices of practical interest, including various types of gears and clock es­ capements. As an example, the video shows the design of a novel forward-reverse mech­ anism which is much simpler than designs found in the literature on mechanism de­ sign. The program has also been tested on design of actual products which are to be marketed in the near future, showing that using the tool can decrease design time for such mechanisms from 6 months to only 1 day.

FAMING is mainly intended as a proof of concept that qualitative physics can indeed provide support for creative design. The combination of case-based reasoining with qualitative physics is applicable to many other domains which may be commercially more promising, such as nanotechnology, drug design or biomedical engineering.