Excitation Monitoring for Online Parameter Estimation

Sufficient excitation is the basis for precise and reliable parameter estimates. Outside of test flights, sufficient excitation often cannot be guaranteed resulting in errors and drift of the estimates. In this study, we present four approaches for monitoring excitation in online parameter estimation, suitable for use with recursive least squares, sequential least squares, extended Kalman filtering, and output-error methods. The excitation monitoring approaches are based on detecting collinearity in the regressor matrix and sensitivity matrices as well as monitoring the magnitude of the sensitivities. We integrated the monitoring into the joint extended Kalman filter and the sequential least squares in frequency-domain. By this, we mitigate drift and noise in the estimates while maintaining the tracking capabilities for changing parameters. We compared these approaches and demonstrated their effectiveness in simulation of longitudinal aircraft dynamics and based on quadcopter flight test data.     «

Sufficient excitation is the basis for precise and reliable parameter estimates. Outside of test flights, sufficient excitation often cannot be guaranteed resulting in errors and drift of the estimates. In this study, we present four approaches for monitoring excitation in online parameter estimation, suitable for use with recursive least squares, sequential least squares, extended Kalman filtering, and output-error methods. The excitation monitoring approaches are based on detecting collinearit...     »