| Modern control design is commonly accompanied by the challenge of dealing with complex systems or plants that combine nonlinear behavior, state constraints, parametric uncertainty, and time varying characteristics. Because of these features, such systems are often difficult to model using first principles, and as a result, the task of designing a suitable controller becomes difficult.; One simple approach to circumvent this problem is to use a calibration map. The purpose of the map is to translate desired outputs or desired states into appropriate inputs for the plant. Usually, the map is obtained from the inversion of the steady state characteristics of the plant, and is stored in a lookup table. The output of the lookup table is then employed for feedforward compensation in open-loop or closed-loop control. However, when the plant is time varying, the resulting control performance is affected by the fact that the calibration becomes less reliable as time progresses.; In an effort to present a new alternative, this research develops a general theoretical framework for online auto-calibration and control of general nonlinear systems. More specifically, the inverse input-state mapping of the plant is learned while in operation (online). This is done by employing a simple neural network structure that emulates an adaptive lookup table. In this approach, the inverse mapping is learned from the current and past states and it is refined in a composite manner by employing input errors and state errors. The learned mapping is used simultaneously in the feedforward path to control the plant along the desired state trajectory. The performance of the plant subject to this novel controller is verified through simulations and experimental data. It is seen that good state tracking is achieved without requiring exact or detailed knowledge of the plant. The main requirements for the successful operation of the novel control approach are the knowledge of the order of the plant and some generic data to initialize the inverse mapping. This last requirement can be fulfilled from steady-state data.; The control theory developed herein is applied to a novel Electro-Hydraulic Poppet Valve (EHPV), which is currently controlled open-loop via an inverse calibration map. The EHPV's are used in a Wheatstone bridge arrangement for motion control of hydraulic actuators. Such a configuration is preferred over the conventional use of spool valves due to the energy savings potential. It is shown in this dissertation that this method improves the value of using these types of valves for motion control in hydraulics. This is due to the combination of self-learning (auto-calibration) and better performance, which results in a more efficient operation of hydraulic equipment. Additionally, it is shown that the auto-calibration of the valves can be used for health monitoring of the same, which consequently improves their reliability, expedites maintenance downtime, and reduces maintenance costs. |
