EAPSI: Generating Models for Systems Based on Their Response to Unmeasured Inputs

Estes Aaron E, Amherst NY

Investigators

Abstract

Automatic control permeates modern life--from the autonomous navigation of ground and air vehicles to the regulation of glucose levels in the bloodstream via an artificial pancreas; but the quality of controllers stems from an intricate understanding of the underlying processes they manipulate. System identification is the technique of building a mathematical model for a dynamic process by observing its reaction to different inputs. The identified model uncovers the fundamental interactions at play in the system, and helps inform the design of controllers. However, the inputs creating a system response are often uncertain, or unmeasured; such as the turbulent forces buffeting an aircraft during flight. This research will develop techniques for generating a model for a system when its response is observed, but the input causing this response is unknown. This research will be conducted at National Cheng Kung University in Tainan City, Taiwan, in collaboration with Dr. Jer-Nan Juang, a pioneer in the field of system identification. The popular Eigensystem Realization Algorithm (ERA) will be extended to identify models for linear time-invariant systems where only output data is available. Specifically, the outcomes include: 1) the rigorous definition of the set of conditions a system must satisfy in the presence of input uncertainty in order to be identifiable; including the locations of sensors and/or actuators, and the relative modal content of the system and the input; 2) an augmentation of the ERA algorithm permitting the identification of plants and inputs, down to a constant scaling factor, from output-only data; and 3) a linear solution for the recovery of the transformation which maps the identified system into physical coordinates, allowing for the decoupling of system modes and input modes. This NSF EAPSI award is funded in collaboration with the Ministry of Science and Technology in Taiwan.