High-level Feedback Control With Neural Networks

Complex industrial or robotic systems with uncertainty and disturbances are difficult to control. As system uncertainty or performance requirements increase, it becomes necessary to augment traditional feedback controllers with additional feedback loops that effectively "add intelligence" to the system. Some theories of artificial intelligence (AI) are now showing how complex machine systems should mimic human cognitive and biological processes to improve their capabilities for dealing with uncertainty.This book bridges the gap between feedback control and AI. It provides design techniques for "high-level" neural-network feedback-control topologies that contain servo-level feedback-control loops as well as AI decision and training at the higher levels. Several advanced feedback topologies containing neural networks are presented, including "dynamic output feedback", "reinforcement learning" and "optimal design", as well as a "fuzzy-logic reinforcement" controller. The control topologies are intuitive, yet are derived using sound mathematical principles where proofs of stability are given so that closed-loop performance can be relied upon in using these control systems. Computer-simulation examples are given to illustrate the performance.