Application of Experience Mapping based Predictive Controller (EMPC) for Under-damped and Unstable Systems

Abstract

Experience Mapping based Predictive Controller (EMPC) is a concept based on the principle of Human Motor Control, that was earlier developed and applied to control a well damped Type 1 system. In this thesis, the concepts of EMPC have been expanded and applied to control an under-damped Type 1 system to achieve reduced overshoots and oscillations. The proposed controller is applied to a DC motor based positioning system with a load coupled through a flexible shaft, which constitutes an under damped position system. EMPC uses the concept of learning by experience and generates an Experience Mapped Knowledge (EMK) which stores a one-to-one mapping of the control parameter to the corresponding steady state value of the parameter to be controlled. The EMK is generated by applying various control actions to the system with different values of the control parameter and corresponding steady state values are recorded. EMK helps EMPC to give the right control action for a given demand by using linear interpolation method. Simulation and practical experimental results show that the proposed controller performs better than traditional controllers like the Proportional-Derivative (PD), and State Space based controllers like the Linear Quadratic Regulator (LQR) and the Linear Quadratic Gaussian (LQG) controller. Stability of EMPC in the presence of non-linearities and various changes in system parameters such as dry friction, actuator saturation, load inertia and spring constant and adaptability of the controller for the same are also discussed with suitable simulation results. The concepts of EMPC are further modified to suit systems containing Backlash as an example. EMPC demonstrates reduced overshoots and zero steady state error in both simulation and practical system. EMPC is practically applied to control an inverted pendulum which does balancing and centring of the carriage simultaneously