Evolving a lookup table based motion controller for a ball-plate system with fault tolerant capabilites

Abstract

This paper describes a novel motion controller, based on evolved lookup tables that have been developed to move a ball to a set-point on the classic ball-plate control system. A fault tolerant controller that could adapt to fault conditions was evolved using a genetic algorithm and evaluated. The three dimensional lookup tables related the current ball position, ball velocity and plate angle to the plate angular velocity required to control the ball. Three sizes of lookup table with different quantization levels of inputs and outputs were evaluated to determine the optimum lookup table size for ball control versus the evolution time. It was found that successful solutions could be evolved within 100 minutes. Fault tolerance was investigated where two position sensor faults were introduced to an evolved solution, and the evolution was restarted with seeded one position sensor fault solutions. It was found that the evolving controller's performance was significantly improved within 1 second, and fully recovered within 10 to 30 seconds.