Helicopter flight control design using a learning control approach

We introduce a neural learning control mechanism for helicopter flight control design. The significance of our contribution is twofold. First neural dynamic programming (NDP) is in its early development stage and successful applications to date have been limited to simple systems, typically those possessing only a single control and a handful of states. With our industrial scale helicopter model, we consider a very realistic class of complex design problems. To accommodate such complex systems we introduce the concept of a trim network which is seamlessly integrated into our NDP control structure and is trained using our NDP control structure. Second, we introduce a new class of design methodologies to the helicopter control system design community. This approach is expected to be effective in dealing with real-time learning applications such as reconfigurable control. The paper consists of a comprehensive treatise of NDP and extensive simulation studies of NDP designs for controlling an Apache helicopter. All of our designs are tested using FLYRT, a sophisticated industry-scale non-linear validated model of the Apache helicopter. Though illustrated for helicopters, our NDP control system framework should be applicable to general control systems