Experimental Identification of Modeling Errors in Dynamic Systems

This paper presents a method for identifying the errors of a linear, time-invariant model of a dynamic system. The identification is based on system inputs and outputs recorded during an experiment. The identified errors are expressed as tolerances of the model parameters. These tolerances are such that a simulation of the dynamic system in which the linear model parameters are allowed to vary within their identified error tolerances can reproduce the response recorded during the experiment. The method is based on linear programming which guarantees that the identified errors are the smallest possible which can reproduce the experimental data. The model error information is useful for failure detection and isolation, system simulation and control design. The identification method is illustrated using a flight-dynamics example in conjunction with the Reachable Measurement Intervals (RMI) failure detection method for systems with modeling errors.