Sliding mode robot controller parameter tuning with genetic algorithms and fuzzy logic

Official URL: http://risc01.sabanciuniv.edu/record=b1505294 (Table of Contents)

Sliding Mode Controllers (SMC) possess robustness properties under parameter uncertainties. Usually, a Lyapunov based controller design with a switching control signal constitutes the backbone of robustness. However, the ideally zero switching time of the controller output cannot be achieved in digital implementation. This causes a phenomenon called chattering – high frequency oscillations observed in systems state variables. Chattering also shows itself as high amplitude oscillatory behavior in the control signal. A chattering actuator output is not favorable for many plants, including robot manipulators driven by actuator torques. This problem is traditionally solved by smoothing the switching control output, deviating from the original mathematical foundations robustness. Over-smoothing causes performance deterioration, while too limited smoothing action may lead to the wear of the mechanical system components. This motivates the exploration of automatic tuning approaches which consider chattering and performance simultaneously. This thesis proposes two SMC smoothing and parameter tuning methods with soft computing (SC) methodologies. The first method is based on Genetic Algorithms (GA). SMC controller parameters, including the ones governing the smoothing action are tuned off-line by evolutionary computing. A measure is employed to assess the instantaneous level of chattering. The integral of this value combined with performance indicators including the rise time and steady state error in a step reference scenario are used as the fitness function. The method is tested on the model of a direct drive (DD) SCARA type robot, via simulations. The GA-tuned SMC is, however, tailored for a fixed reference signal and fixed payload. Different references and payload values may pronounce the chattering effects or lead to performance loss due to over-smoothing. The second SMC parameter tuning method proposed employs a fuzzy logic system to enlarge the applicability range of the controller. The chattering measure and the sliding variable are used as the inputs of this system, which tunes the controller output smoothing mechanism on-line, as opposed to the off-line GA technique. Again, simulations with the direct-drive robot model are employed to test the control and tuning method.