| Physical parameter identification is useful in many applications, especially in aerospace where much analysis goes into developing accurate physical system models for control. A number of off-line physical parameter identification methods exist; however, the choice of on-line methods is more limited. On-line identification methods are required for adaptive control. New on-line physical parameter identification methods are developed in this work as motivated by the problem of launch vehicle adaptive control.; Launch vehicles vary from launch to launch due to differences in payloads and fuel loading. Based on the known variations, launch vehicle control laws are re-analyzed and modified if necessary; this process is expensive and adds to recurring launch vehicle costs.; This reanalysis is performed despite the fact that changes in the launch vehicle are relatively minor. A trustworthy adaptive control system could eliminate this expensive redesign cycle. An adaptive control system could also provide better performance than a controller redesigned off-line. However, adaptive control is still considered too risky to use with unstable systems, primarily due to limitations in the identification methods currently available for use in adaptive control. This problem is addressed with the development of new identification algorithms.; A philosophy of identification is described which uses physical parameters for identification. A technique is developed to convert existing on-line methods to a form capable of identifying physical parameters. New methods include Physical Parameter versions of Normalized Least Mean Squares (NLMS), Recursive Least Squares (RLS), Extended Least Squares (ELS), Recursive Maximum Likelihood (RML), and the Extended Kalman Filter (EKF). Compared to transfer function identification, physical parameter identification reduces the order of the problem and speeds up convergence. Compared to the Extended Kalman Filter, the new methods have a faster iteration time. The use of physical parameters also allows bounding based on expected parameter variations.; The new identification methods are compared with existing methods and demonstrated in simulations of a two-mass/spring system, an OH-6A helicopter near hover, and the National Launch System 1.5-Stage launch vehicle. |
