High Performace Gun-Pointing Control Using Advanced Nonlinear and Robust Techniques

In applications such as tank gun-pointing control, neural nets are generally used to learn the forward or inverse non-linear dynamics of the plant being controlled. Once the neural net has learned the plant dynamics, this dynamics can be used for feedforward or feedback control in the conventional manner. Aside from the important issue of stability, this prior art required that the neural net be trained with "training pairs." It is necessary to know the desired output of the system for many inputs so that the neural net can learn the plant dynamics. This is called supervised learning, and in many cases it is not possible to generate such training pairs. In this proposal, we propose to use a hybrid combination of neural net controller with H-infinity control which can guarantee stability. The role of neural net is to compensate the nonlinear dynamics and the purpose of H-infinity control is to ensure the robustness with respect to high frequency unmodeled dynamics, actuator saturation, and sensor noises. A side benefit of this approach is that conventional training pairs are not needed. Learning can be unsupervised. Yet another advantage of our approach is that the proposed neural net controller is reusable within a class of non-linear system. Our proposed method remains the same irrespective of the changes in tank system parameters such as barrel length and mass. A key issue in developing an intelligent control system is the reusability of the low-level controller, i.e. the same controller works even if the behavior of the system or the parameters of the system have been changed. This is the case for the controller proposed here. The hybrid combination of neural network and H-infinity control is novel in tank gun-pointing control. We can apply the method to other systems such as process control, motor control, missile control, flight control, engine control, automobile control, flight control, and precision instrumentation systems.