Reasoning about Sensor Data for Automated System Identification

Article type: Research Article

Authors: Bradley, Elizabeth^{; 1} | Easley, Matthew^{; *}

Affiliations: University of Colorado Department of Computer Science, Boulder, CO 80309-0430, USA

Correspondence: [*] Corresponding author. E-mail: [email protected].

Note: [1] E-mail: [email protected].

Abstract: 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.

Keywords: System identification, Qualitative physics, Automated sensor/actuator interaction, Delay-coordinate embedding

DOI: 10.3233/IDA-1998-2205

Journal: Intelligent Data Analysis, vol. 2, no. 2, pp. 123-138, 1998