Iterative learning control of an electrohydraulic injection molding machine with smoothed fill-to-pack transition and adaptive filtering

The general focus of this research is the development of novel control algorithms and their application to electrohydraulic systems. The methodologies presented are appropriate for many types of industrial applications that have decidedly nonlinear dynamical characteristics. In particular, this work demonstrates the utilization of these advanced algorithms on an Injection Molding Machine (IMM) that is driven by an electrohydraulic actuation system. The topics covered by this research involve modeling, theoretical analysis, and experimental implementation.; A physics-based IMM model is developed starting with first-principles. This model is utilized throughout the work as a test-bed for controller design and tuning. The model itself is a valuable contribution since it supercedes previous modeling attempts that focused on either the polymer process or the machine. By incorporating both the machine components and the process loading, a more accurate overall model is developed that is applicable over the entire injection profile. This model is able to emulate the behavior of the filling and packing phases as well as the fill-to-pack transition.; The control algorithms used in this work belong to the class of Iterative Learning Control (ILC) approaches. These are very effective for the control of industrial processes that repeat themselves, such as injection molding. The state-of-the-art in ILC is extended by a novel adaptive filtering approach explicitly designed to handle nonsmooth nonlinearities. These types of nonlinearities appear often in electro-hydraulic systems. The aforementioned adaptive filtering approach is analyzed and proven to be convergent for a specific class of dynamical systems.; The control algorithms are implemented and tested on an industrial IMM as well as a smaller physical testbed. The results of the experimental investigations demonstrate several things. The system model is shown to be very accurate and provides an excellent simulation environment for rapid controller design and tuning. The ILC approaches are shown to provide a significant increase in system performance. In addition to the learning controllers, a Bumpless Transfer approach is also utilized to smooth the transition between the separate injection phases.