Non parametric model reference adaptive control

This thesis introduces a new method of adaptive control: NonParametric Model-Reference Adaptive-Control (NP-MRAC). The method relies on estimating the shape of the Finite Impulse Response (FIR) of the plant and the FIR of the controller rather than determining the plant and the controller parameters. An FIR adaptive filter was used to identify the FIR of the plant and calculate the FIR of the controller. The NP-MRAC was studied both by digital simulation on the NEC-APC-3 computer and by real-time hybrid-simulation on the TMS320C20 signal processor. In this thesis, the NP-MRAC employed the LMS algorithm as an adaptive algorithm. This algorithm is simple, yet performs satisfactorily. The NP-MRAC, using the LMS algorithm, has excellent adaption capabilities in the presence of both parameter variations and time delay variations. It also allows for closed-loop pole placement and because there are no restrictions on closed-loop zeros, good set-point tracking can be achieved by a suitable choice of the reference-model. The scheme's major limitations are relatively slow rate of convergence and sensitivity to variations in the eigenvalue spread, defined as the ratio of the maximum to minimum eigenvalue of the correlation matrix of the FIR adaptive filter input signal. However the NP-MRAC should have great potential in applications where time-delay and parameter variations are relatively gradual, as it is generally known that FIR filters are more robust than their recursive Infinite Impulse Response IIR counterpart.