L 1 Adaptive Output - Feedback Control Architectures by Evgeny Kharisov

This research focuses on development of L1 adaptive output-feedback control. The objective is to extend the L1 adaptive control framework to a wider class of systems, as well as obtain architectures that afford more straightforward tuning. We start by considering an existing L1 adaptive output-feedback controller for nonstrictly positive real systems based on piecewise constant adaptation law. It is shown that L1 adaptive control architectures achieve decoupling of adaptation from control, which leads to bounded away from zero time-delay and gain margins in the presence of arbitrarily fast adaptation. Computed performance bounds provide quantifiable performance guarantees both for system output and control signal in transient and steady state. A noticeable feature of the L1 adaptive controller is that its output behavior can be made close to the behavior of a linear time-invariant system. In particular, proper design of the lowpass filter can achieve output response, which almost scales for different step reference commands. This property is relevant to applications with human operator in the loop (for example: control augmentation systems of piloted aircraft), since predictability of the system response is necessary for adequate performance of the operator. Next we present applications of the L1 adaptive output-feedback controller in two different fields of engineering: feedback control of human anesthesia, and ascent control of a NASA crew launch vehicle (CLV). The purpose of the feedback controller for anesthesia is to ensure that the patient’s level of sedation during surgery follows a prespecified profile. The L1 controller is enabled by anesthesiologist after he/she achieves sufficient patient sedation level by introducing sedatives manually. This problem formulation requires safe switching mechanism, which avoids controller initialization transients. For this purpose, we used an L1 adaptive controller with special output predictor initialization routine, which provides bumpless transient during switches. For the second application, our objective was to design a single controller without parameter scheduling, which would cover the whole flight envelope of the first stage of the CLV. This approach has the potential of reducing the design costs by reducing the number of necessary wind tunnel tests. One of the main challenges we encountered was variability of the parameters of the CLV. Both aerodynamic and inertia parameters change dramatically during the CLV operation. The fact that CLV inertia significantly reduces with time allows for demanding faster controller response and more agile CLV behavior as time flows. This inspired us to develop an L1 adaptive controller, which would take into account for changes in the desired control specifications without the need for switching the control laws. This