Disturbance estimation and parameter identification algorithms for vehicle systems.

This dissertation develops several estimation schemes to identify unknown states, parameters and disturbances for both linear and nonlinear systems. Two types of algorithms are studied: (i) algorithms to estimate unknown parameters and/or states (adaptive observer algorithms); and (ii) algorithms to estimate external disturbances and/or states (disturbance observer algorithms). The adaptive observer algorithms developed in this dissertation are applicable to systems which are linearly or nonlinearly dependent on the unknown parameters. For the systems which are linear in the parameters, the proposed schemes are classified into full-state feedback approaches and output feedback approaches. The full-state feedback observers are derived from Lyapunov design techniques. The output feedback observers can be derived from either the least squares method or the Lyapunov approach. In contrast to previous works, the physical sense of estimated states and parameters is preserved in the proposed output feedback schemes, since we constructed the algorithms from physical systems instead of canonical forms. For the systems that are nonlinearly dependent on the unknown parameters, the observer needs full-state feedback, and the Lyapunov approach is used. We have also studied the cases in which the unknown parameters are correlated and we have found that the estimation is noticeably improved by properly using knowledge of this correlation. The disturbance observer algorithms developed in this dissertation are also divided into full-state feedback and output feedback approaches. For the full-state feedback approaches, a feedback correction term is added to the estimation so that better estimation performance is obtained when disturbances are slowly varying. For the output feedback case, we apply inverse dynamics to construct the identification schemes. The disturbance and state estimation errors are shown to converge exponentially to zero. Two modified estimation schemes are proposed for linear nonminimum phase systems. The proposed algorithms were motivated by our study of vehicle control problems. The proposed methods are applied to several vehicle control examples, including the estimation of vehicle parameters, external disturbances (road super-elevation and wind gusts), tire forces, etc. In all these numerical studies, the proposed methods have demonstrated satisfactory performance.Ph.D.Applied SciencesAutomotive engineeringMechanical engineeringSystems scienceUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/130295/2/9722031.pd