New adaptation methodologies applied to adaptive control systems and system identification

In this research we propose a new self-tuning or adaptation algorithm based on a theory of adaptive interaction. The application is for self-tuning of PID controller, adaptation of weights for neural networks used for nonlinear function identification and neural network controller. The adaptation algorithm is also used for system identification. The theory develops a simple and effective way to perform gradient descent in the parameter space. The approximation version of the algorithm requires no knowledge of the plant to be controlled. This makes the algorithm robust to changes in the plant. It also makes the algorithm universally applicable to linear and nonlinear plants. In all the investigated configurations, the algorithm achieves the tuning objective by minimizing an error function. The self-tuning PID controller was used with second and third-order plants, stable and unstable plants, linear and nonlinear plants and nonminimum phase plants. Because of its simplicity, the computing overhead for adding the self-tuning algorithm is negligible. We applied the self-tuning PID controller in an automotive sensor manufactured by Hitachi to satisfy performance requirements often conflicting for warm-up, response time, steady state error and overall robustness. Successful results are demonstrated. For the neural network application, the tuning algorithm is used for function approximation and neural network controller. Unlike the approach based on the well-known backpropagation algorithm, this approach will not require the controlled plant to be converted to its neural network equivalent, a major obstacle in early approaches. By applying the adaptation algorithm, the same adaptation as the backpropagation algorithm is achieved without the need of backward propagating the error throughout the feedback network. This important property makes it possible to adapt the neural network controller directly. The neural network controller was applied to minimum and nonminimum phase plants. For the dynamic system identification, a variation of the adaptive interaction algorithm is used to update the parameter of the system. The system would be in transfer function form, in observable canonical form or in state space representation. These approaches allowed identification of second and third order systems.