Simultaneous State and Parameter Estimation Study

Overview

Here I'm just linking to a paper I wrote on simultaneous state and parameter estimation using an aircraft model. I intend to continue the questions explored here on the actual aircraft.

Summary of Paper

One of the most difficult aspects of predicting aircraft dynamic behavior is obtaining accurate values for the aerodynamic coefficients. Though some analytical tools and simulations are available, often the simplest and most accurate method is through parameter estimation based on the actual flight data. For real-world systems, the true aircraft dynamics are never perfectly known and instead are only reported through sensor measurements. Therefore, in order to learn the aerodynamic parameters, both the parameters and the aircraft’s dynamics must be simultaneously estimated. Through an extended Kalman filter (EKF) the full state was established to estimate both the states describing the aircraft’s dynamics and the states describing its parameters. The EKF performed estimation on the full state through a short maneuver. After the maneuver was complete, the states were passed into a Kalman smoother (KS) in an attempt to enhance the estimates generated by the EKF. Experimentation was performed with regard to the effect that the number of unknown parameters had on estimation performance. In addition, the ability of the KS to improve both estimates of the dynamics states and parameters was explored. The KS decreased the estimate error by 59% for the dynamics states but made an insignificant difference to the parameter estimates. The parameter estimates themselves decreased in accuracy as the number of unknown parameters increased but many were still capable of convergence on their true value.

Full Paper