Sundararajan N., Saratchandran P., Li V. Fully Tuned Radial Basis Function Neural Networks for Flight Control

Springer, 2002. — 167 p.In the past three decades, major advances have been made in adaptive control of linear time-invariant plants with unknown parameters. The choice of the controller structure is based on well established results in linear systems theory, and stable adaptive laws which assure the global stability of the overall systems are derived based on the properties of those systems. In contrast this, mature design procedures that simultaneously meet the requirements of stability, robustness, and good dynamic response for nonlinear system control are currently not available.
Recently, Artificial Neural Network (ANN) based control strategies have attracted much attention because of their powerful ability to approximate continuous nonlinear functions. Specifically, a neural controller with on-line learning can adapt to the changes in system dynamics and hence is an ideal choice for controlling highly nonlinear systems with uncertainty. Among a variety of network structures, Radial Basis Function Network (RBFN) has been studied intensively due to its good generalization ability and a simple network structure that avoids unnecessary and lengthy calculations. All the advantages of the RBFN have motivated us to further investigate its use in the area of nonlinear adaptive control in this book, with emphasis in aircraft flight control applications.
The classical approach for Gaussian RBFN implementation is to fix the number of hidden neurons, centers and widths of the Gaussian function a priori, and then estimate the weights connecting the hidden and output layers using parameter tuning rules, like LMS, RLS etc. However, in practice it is difficult to choose the centers and widths appropriately, especially for on-line implementation where preliminary training is impossible. The inaccurate centers and widths will unavoidably result in the deterioration of the performance, especially when coping with highly nonlinear systems with uncertainty, such robot, aircraft, etc. In comparison to conventional approaches, recently fully tuned RBFNs have shown their potential for accurate identification and control. In a fully tuned RBFN, not only the weights of the output layer, but also the other parameters of the network (like the centers and widths) are updated, so that the nonlinearities of the dynamic system can be captured as quickly possible.
In this book, we first address the theoretical aspects of designing stable nonlinear adaptive control law with a fully tuned RBFN, and then explore the applications of the controllers designed for aircraft flight control. More specifically, the objectives of the book can be summarized as:
To design indirect adaptive control and direct adaptive control strategies incorporating fully tuned RBFN networks. In the indirect control strategy, a stable identification scheme using the fully tuned RBFN is developed for identification of nonlinear systems with external inputs. In the direct adaptive control scheme, the objective is to design the on-line control law based on a fully tuned RBFN, guaranteeing the stability of the overall system.
To explore the applications of the proposed neuro-controller in the field aircraft flight control. Simulation studies are carried out based on different control objectives and aircraft models, including command following for a linearized F8 aircraft model in longitudinal mode, pitch-rate control for a localized nonlinear fighter aircraft model, and implementing a high stability-axis roll maneuver based on a full-fledged 6-DOF high performance aircraft model with nonlinear dynamic nature.
To evaluate the recently developed MRAN algorithm for real-time nonlinear system identification and adaptive control, especially in fault tolerant aircraft flight control applications.A Review of Nonlinear Adaptive Neural Control Schemes
Part I Nonlinear System Identification and Indirect Adaptive Control Schemes
Nonlinear System Identification using Lyapunov-based Fully Tuned RBFN
Real-Time Identification of Nonlinear Systems using MRAN/EMRAN Algorithm
Indirect Adaptive Control using Fully Tuned RBFN
Part II Direct Adaptive Control Strategy and Fighter Aircraft Applications
Direct Adaptive Neuro Flight Controller using Fully Tuned RBFN
Aircraft Flight Control Applications using Direct Adaptive NFC
MRAN Neuro-Flight-Controller for Robust Aircraft Control
Conclusions and Future Work