Design of a rule-based controller for improved electrodynamic levitation system performance.

Mankind's willingness for rapid travel has always powered relentless research in transportation. This led to the persistent revolution over a few decades that made it possible for humanity to safely move from one point to another on the planet in a matter of hours. Electrodynamic levitation technology has seen increasing interest over the years for the possibility of being employed in a new revolution of public transportation. Hyperloop TT has envisioned a maglev technology based on passive magnets arranged in a Halbach array arrangement. The use of only passive components renders the concept particularly appealing from the energy consumption and the reliability standpoint. Furthermore, the use of Hyperloop capsules inside near-vacuum tubes allows for nearly frictionless movement and very high speeds. However, electrodynamic levitation shows inherent instability when used to levitate a single degree of freedom system. Instability problems are solved by separating the levitated body into two masses. Regardless, external excitations such as ground irregularities or even changes in track slope may result in unwanted oscillatory behavior that, if not controlled, may result in an unpleasant ride for passengers. In this context, this thesis work aims to investigate the dynamic identity of an electrodynamically levitated system by resorting to a quarter car model in order to study its stability, and the effects that variations in the major parameters produce on its dynamic behavior. Subsequently, the main focus is to improve passenger ride quality by developing a controller based on fuzzy logic, proving the effectiveness of such control approach on a levitated system. Control will be materially exerted by a voice coil actuator whose quickness in responding to rapidly changing input signals renders it particularly well suited for the job. Conventional fuzzy logic controllers are designed to work on a specified range of inputs which is rarely supposed to exceed predefined bounds. In a real context, however, given the possibility for random inputs to occur, it is not feasible to solely rely on an estimated range of variation of inputs and it is even less feasible to define fuzzy membership functions over a very broad range as the number of rules would grow exponentially. It is for this reason that a novel approach to fuzzy logic control is presented, which involves real-time estimation of the magnitude of the phenomena in order to dynamically re-scale the rules associated with the fuzzy controller and ensure a consistent and constant level of control resolution. The extensive use of lookup tables and the necessity for small sets of rules paves the way to heavy optimization. Moreover, working with normalized fuzzy control spaces renders it possible to develop an algorithm to train a fuzzy inference system starting from the behavior of another one, essentially translating the actions taken by the training system into rules based on different input states. This approach becomes particularly convenient when only a restricted number of states are known in real-time and when they do not allow for an easy and reliable definition of fuzzy rules. The information provided to the fuzzy controller is enriched by state estimations obtained from a Kalman filter that has been optimized to operate with a reduced form of the mathematical model representing the system.