Intelligent control of integrated cubic neural network for robustness and fault-tolerance

This study aims at establishing a robust intelligent control method with higher control performance and wider applicable region by extending the Cubic Neural Network (CNN) intelligent control method which consists of multilevel parallel processing on different degrees of abstraction. In this study, an integrated CNN (ICNN) intelligent control method for nonlinear systems is proposed. In the method, an integrator neural network acquires suitable switching of several CNN controllers for a different local objective based on an evaluation of system states. the proposed ICNN is applied to a control problem of a swung up and inverted pendulum mounted on a cart for the case that arbitrary initial condition of pendulum angle. In order to confirm the performance of the ICNN, computer simulations and experiments using a real apparatus were carried out for the cases of parameter variation and sensor failure. As a result, it was demonstrated that the ICNN controller can stand up the pendulum taking into account the cart position limit at abnormal situations. Then, the robustness and the fault-tolerance of the ICNN were verified in comparison with the sliding mode control technique.