Reasoning about Sensor Data for Automated System Identification

The computer program PRET automatically constructs mathematical models of physical systems. A critical part of this task is automating the processing of sensor data. PRET's intelligent data analyzer uses geometric reasoning to infer qualitative information from quantitative data; if critical variables are either unknown or cannot be measured, it uses delay-coordinate embedding to reconstruct the internal dynamics from the external sensor measurements. Successful modeling results for two sensor-equipped systems, a driven pendulum and a radio-controlled car, demonstrate the effectiveness of these techniques.