Plant-friendly input signal design for System Identification and robust control performance

System Identification is associated with the techniques for construction of adequate mathematical models for an unknown dynamical system from available input-output data. The quality of the collected data is critical in developing a model that accurately represents the true system for the intended application. The data collection process is often iterative since a priori information about the plant may be limited and the input signal may need to be redesigned each iteration. This process presents input signal design challenges for model development that have been addressed in this dissertation.;To support this framework, a modified Simultaneous Perturbation Stochastic Approximation (SPSA) methodology (MSPSA-K) was developed to generate plant-friendly input signals where the input signal may have an arbitrarily defined spectrum (in both amplitude and frequency spacing). This methodology provides a flexible computational framework to incorporate objective functions with various plant-friendly performance measures and allows for tradeoffs between the different measures. A generalized probabilistic uncertainty description was derived based on an input signal composed of a sequence of multi-sinusoidal signals, where each signal in the sequence may have a different spectral shape, number of periods, and fundamental frequency. This uncertainty description enables the use of adaptive algorithms for input signal adjustment during experimental testing in system identification to obtain better plant estimates, shape the probabilistic uncertainty description, and to reduce the input signal duration.;The algorithms developed from this research are easy to implement and are not computationally intensive and the framework can easily be extended to incorporate additional features and capabilities. The practical importance and significance of this work is clearly that the identification and control algorithms can be used for robust control development in many contexts including process control, aerospace, and identification test monitoring.;To address these issues a comprehensive framework was developed that incorporates the generation of plant friendly input signals, computation of an aggregate Empirical Transfer Function Estimate (ETFE) plant estimate, computation of a probabilistic uncertainty description, and adaptive adjustment of the input signal spectrum with the objective of satisfying the robust performance criteria using an input signal sequence composed of multi-sinusoidal signals.