Marcello R. Napolitano

A fault tolerant flight control system for sensor and actuator failures using neural networks

Abstract In recent years neural networks have been proposed for identification and control of linear and non-linear dynamic systems. This paper describes the performance of a neural network-based fault-tolerant system within a flight control system. This fault-tolerant flight control system integrates sensor and actuator failure detection, identification, and accommodation (SFDIA and AFDIA). The first task is achieved by incorporating a main neural network (MNN) and a set of n decentralized neural networks (DNNs) to create a system with n sensors which has the ability to detect a wide variety of sensor failures. The second scheme implements the same main neural network integrated with three neural network controllers. The contribution of this paper focuses on enhancements of the SFDIA scheme to allow the handling of soft failures as well as addressing the issue of integrating the SFDIA and the AFDIA schemes without degradation of performance in terms of false alarm rates and incorrect failure identification. The results of the simulation with different actuator and sensor failures with a non-linear aircraft model are presented and discussed.

Kalman filters and neural-network schemes for sensor validation in flight control systems

Detection, identification, and accommodation of sensor failures can be a challenging task for complex dynamic systems. This paper presents the comparison of two different approaches for the task of sensor failure detection, identification, and accommodation in a flight control system assumed to be without physical redundancy in the sensory capabilities. The first approach is based on the use of a set of online learning neural networks; the second approach is based on the use of a bank of Kalman filters. The objective is to evaluate the robustness of both schemes; the comparison is performed through testing of the schemes for several types of failures presenting different level of complexity in terms of detectability. The required computational effort for both schemes is also evaluated. For each of these failure types this comparison is performed at nominal conditions, that is with the system model and its noise perfectly modeled (with the Kalman filter scheme performing at nominal conditions), and at conditions, where discrepancies occur for the modeling of the system as well as the system and measurement noises. While the Kalman-filter-based scheme takes advantage of its robustness capabilities, the neural-network-based scheme, starting from a random numerical architecture, relies on the learning accumulated either online or from off-line simulations. The study reveals that online learning neural architectures have potential for online estimation purposes in a sensor validation scheme, particularly in the case of poorly modeled dynamics.

Neural-Network-Based Scheme for Sensor Failure Detection, Identification, and Accommodation

This paper presents a neural-network-based approach for the problem of sensor failure detection, identification, and accommodation for a flight control system without physical redundancy in the sensors. The approach is based on the introduction of on-line learning neural network estimators. For a system with n sensors, a combination of a main neural network and a set of n decentralized neural networks achieves the design goal. The main neural network and the ith decentralized neural network detect and identify a failure of the ith sensor, whereas the output of the ith decentralized neural network accommodates for the failure by replacing the signal from the failed ith sensor with its estimate. The on-line learning for these neural network architectures is performed using the extended back-propagation algorithm. The document describes successful simulations of the sensor failure detection, identification, and accommodation process following both soft and hard sensor failures. The simulations have shown remarkable capabilities for this neural scheme.

Aircraft failure detection and identification using neural networks

In this paper, a neural network is proposed as an approach to the task of failure detection following damage to an aerodynamic surface of an aircraft flight control system. Several drawbacks of other failure detection techniques can be avoided by taking advantage of the flexible learning and generalization capabilities of a neural network. This structure, used for state estimation purposes, can be designed and trained on line in flight and generates a residual signal indicating the damage as soon as it occurs. From an analysis of the cross-correlation functions between some key state variables, the identification of the damage type can also be achieved. The results of a nonlinear numerical simulation for a damaged control surface are reported and discussed.

Experimental application of extended Kalman filtering for sensor validation

A sensor failure detection and identification scheme for a closed loop nonlinear system is described. Detection and identification tasks are performed by estimating parameters directly related to potential failures. An extended Kalman filter is used to estimate the fault-related parameters, while a decision algorithm based on threshold logic processes the parameter estimates to detect possible failures. For a realistic evaluation of its performance, the detection scheme has been implemented on an inverted pendulum controlled by real-time control software. The failure detection and identification scheme is tested by applying different types of failures on the sensors of the inverted pendulum. Experimental results are presented to validate the effectiveness of the approach.

Machine Vision/GPS Integration Using EKF for the UAV Aerial Refueling Problem

The purpose of this paper is to propose the application of an extended Kalman filter (EKF) for the sensors fusion task within the problem of aerial refueling for unmanned aerial vehicles (UAVs). Specifically, the EKF is used to combine the position data from a global positioning system (GPS) and a machine vision (MV)-based system for providing a reliable estimation of the tanker-UAV relative position throughout the docking and the refueling phase. The performance of the scheme has been evaluated using a virtual environment specifically developed for the study of the UAV aerial refueling problem. Particularly, the EKF-based sensor fusion scheme integrates GPS data with MV-based estimates of the tanker-UAV position derived through a combination of feature extraction, feature classification, and pose estimation algorithms. The achieved results indicate that the accuracy of the relative position using GPS or MV estimates can be improved by at least one order of magnitude with the use of EKF in lieu of other sensor fusion techniques.

On-Line Learning Nonlinear Direct Neurocontrollers for Restructurable Control Systems

This paper describes an innovative approach to the problem of the on-line determination of a control law in order to achieve a dynamic reconfiguration of an aircraft that has sustained extensive damage to a vital control surface. The approach consists of the use of on-line learning neural network controllers that have the capability of bringing an aircraft, whose dynamics can become unstable after a substantial damage, back to an equilibrium condition. This goal has been achieved through the use of a specific training algorithm, the extended back-propagation algorithm (EBPA), and proper selection of the architectures for the neural network controllers. The EBPA has recently shown remarkable improvements over the back-propagation algorithm in terms of convergence time and local minimum problems. The methodology is illustrated through a nonlinear dynamic simulation of a typical combat maneuver for a high-performance aircraft.

AUTONOMOUS AERIAL REFUELING FOR UAVS USING A COMBINED GPS-MACHINE VISION GUIDANCE

The most important factors affecting the performance of a control scheme for Autonomous Aerial Refueling (AAR) for UAVs are the magnitude of the wake effects from the Tanker and the accuracy of the measurements of the UAV-Tanker distance and attitude leading to the docking. The main objective of the effort described in this paper is the implementation of a detailed modeling and simulation environment for evaluating the AAR problem. In particular, a specific control scheme based on a sensor fusion between GPS- based and Machine Vision-based measurements is proposed. Furthermore, the iterative algorithm used for estimating the position of the optical markers has been modified to be robust to a loss of visibility by one or more optical markers during the docking sequence. The paper presents the results of a detailed analysis of the AAR under different scenarios.

Flight-Test Evaluation of Sensor Fusion Algorithms for Attitude Estimation

In this paper, several Global Positioning System/inertial navigation system (GPS/INS) algorithms are presented using both extended Kalman filter (EKF) and unscented Kalman filter (UKF), and evaluated with respect to performance and complexity. The contributions of this study are that attitude estimates are compared with independent measurements provided by a mechanical vertical gyroscope using 23 diverse sets of flight data, and that a fundamental difference between EKF and UKF with respect to linearization is evaluated.

Online Parameter Estimation Techniques Comparison Within a Fault Tolerant Flight Control System

The results of a study where two online parameter identie cation (PID) methods are compared for application within a fault tolerant e ight control system are described. One of the PID techniques is time-domain based, whereas the second is featured in the frequency domain. The time-domain method was directly suitable for the online estimates of the dimensionless aircraft stability derivatives. The frequency-domain method was modie ed fromitsoriginalformulationtoprovidedirectestimatesofthestabilityderivatives.Thiseffortwasconductedwithin the research activities of the NASA Intelligent Flight Control System F-15 program. The comparison is performed throughdynamicsimulationswithaspecie cproceduretomodeltheaircraftaerodynamicsfollowingtheoccurrence of a battle damage/failure on a primary control surface. The two PID methods show similar performance in terms of accuracy of the estimates, convergence time, and robustness to noise. However, the frequency-domain-based method outperforms the time-domain-based method in terms of computational requirements for online real-time applications. The study has also emphasized the advantages of using ad hoc short preprogrammed maneuvers to provideenoughexcitationfollowingtheoccurrenceoftheactuatorfailuretoallowtheparameterestimationprocess.