This dissertation proposes a pair of strategies for model-free, iteratively adaptive 'on-off' switching control of capacitive micro-actuators with limited sensing rates and significant model uncertainty. A desired actuator motion is to be repeated many times, with the ability to adjust the input sequence between movements. These strategies are intended to limit power consumption of a servo control system for capacitively-loaded piezoelectric micro-actuators of autonomous walking micro-robots.;The first strategy is based on a heuristic dynamic search algorithm that iteratively searches for optimal switching instances, while the second uses a stochastic gradient approximation algorithm. The first method uses simple adaptation laws based on known, general feature of the dynamic behavior of the actuators and a small number of measurements taken during each iteration. The second method adjusts switching instances to minimize an objective function using simultaneously perturbed stochastic approximation (SPSA). The SPSA estimates the gradient of the objective function and uses just a single sensor measurement in each iteration of actuator movement.;Both approaches can be implemented with much lower sensing and driving circuit power than would be needed to implement a conventional control structure such as pulse-width-modulation (PWM) or analog driving with real-time feedback. Energy savings are verified by comparison to the conventional control structure, using power estimates for the components of the servo system. Methods of convergence analysis and conditions are also proposed for predicting convergence speed and stability of both approaches, and applied to sample target piezoelectric actuators. In addition, the controllers are tested in simulation and experimentally on various piezoelectric actuator test beds, and related to future research needs for realization of highly-mobile micro-robots.
展开▼
