Automated reasoning about machines

Abstract

Numerical simulation is often used in predicting machine behavior, a basic capability for many tasks such as design and fault diagnosis. However, using simulators requires considerable human effort both to create behavioral models and to analyze and understand simulation results. I describe algorithms which automate the kinematic and dynamical analysis needed to create behavioral models and which automate the intelligent control of computational simulations needed to understand a machine's behavior over both short and long time scales. The input is a description of a machine's geometry and material properties, and the output is a behavioral model for the machine and a concise qualitative/quantitative prediction of the machine's long-term behavior. My algorithms have been implemented in a working program which can predict a machine's behavior over both short and long time periods. At present this work is limited to mechanical devices, particularly clockwork mechanisms.