| Characterized by its component working at different frequency range, multi-scale systems commonly occur in industry. Their dynamics are often nonlinear, coupled and stiff. Characteristics of such dynamics have to be clearly understood and insights developed so that they can be carefully incorporated into the control law, for reliable and effective designs. Since there are no general guidelines for nonlinear system control, the problem is studied on a case by case basis. This dissertation specifically focuses on the development of robust adaptive learning algorithms for multi-scale systems including issues such as decoupling, model order reduction and control, utilizing insights into the dynamic characteristics. In this dissertation, two example cases are considered, an ODE system (an indexing valve plate industrial grade pump) and a PDE system (a flexible two-link manipulator). Specifically, for the ODE Example Case, since system parts operate in different frequency bandwidths, and high frequency states have periodic characteristics, averaging, perturbation and a novel linearization scheme are used for model reduction. Using insights, a decoupling technique and a learning calibration technique are proposed for control design. For the PDE Example Case, the full model is carefully reorganized using insights and an energy-based stable decoupling control law design is proposed and validated. Learning is also validated here for achieving complete decoupling. |
