Redouane Hallouzi

Fault-Tolerant Subspace Predictive Control Applied to a Boeing 747 Model

This paper presents a fault-tolerant control system based on a combination of predictive control and subspace identification called subspace predictive control. In this control method, the mathematical model used by conventional predictive controllers to predict the future output is replaced by a subspace predictor. Because the subspace predictor is continuously updated in a closed-loop setting based on new input-output data, it can naturally adapt the controller after a fault has occurred. This property is very useful for fault-tolerant control because faults might be unanticipated. A novel feature of the presented subspace predictive control algorithm is that the predictor is recursively updated in a computationally efficient way. A multiple-model fault classification scheme is used to distinguish between anticipated and unanticipated faults. For anticipated faults the control system is configured such that it can accommodate these faults more quickly. The proposed fault-tolerant control system is evaluated on a detailed model of a Boeing 747.

FAULT DETECTION AND IDENTIFICATION OF ACTUATOR FAULTS USING LINEAR PARAMETER VARYING MODELS

Abstract A method is proposed to detect and identify two common classes of actuator faults in nonlinear systems. The two fault classes are total and partial actuator faults. This is accomplished by representing the nonlinear system by a Linear Parameter Varying (LPV) model, which is derived from experimental input-output data. The LPV model is used in a Kalman filter to estimate augmented states, which are directly related to the faults. Decision logic has been developed to determine the fault class from the estimated augmented states. The proposed method has been validated on a nonlinear simulation model of a small commercial aircraft.

Fault Tolerant Flight Control - A Survey

Nowadays, control systems are involved in nearly all aspects of our lives. They are all around us, but their presence is not always really apparent. They are in our kitchens, in our DVD-players, computers and our cars. They are found in elevators, ships, aircraft and spacecraft. Control systems are present in every industry, they are used to control chemical reactors, distillation columns, and nuclear power plants. They are constantly and inexhaustibly working, making our life more comfortable and more efficient...until the system fails.

MODEL WEIGHT AND STATE ESTIMATION FOR MULTIPLE MODEL SYSTEMS APPLIED TO FAULT DETECTION AND IDENTIFICATION

Abstract A method is proposed for estimating both the weights and the state of a multiple model system with one common state vector. In this system, the weights are related to the activation of each individual model. For the resulting nonlinear estimation problem a method is developed that efficiently decomposes the total problem into two separate parts, one for estimating the model weights and one for estimating the state. The method has been validated on a component, actuator and sensor fault detection and identification problem for a linearized model of an aircraft.

MODEL WEIGHT ESTIMATION FOR FDI USING CONVEX FAULT MODELS

Abstract A method is proposed for modeling a large number of faults in a system by a convex combination of a limited number of fault models that form a model set. The fault models in this model set, correspond to the maximum and minimum expected faults for faults that can occur partially. In this way partial faults can be represented by a convex combination of the models from the model set. The identification of faults is performed by estimating the weights of the models from the model set. A set of linearized models of a Boeing 747 aircraft is used to display the effectiveness of the proposed method. This model set also includes models of the aircraft that correspond to faults that occurred during the disastrous crash of EL AL flight 1862 in 1992.

Communication Based Longitudinal Vehicle Control Using an Extended Kalman Filter

This paper presents the design of a longitudinal controller for a cluster of vehicles with inter-vehicle communication (IVC). By applying IVC a smooth traffic flow can be realized. The proposed controller can actively control the throttle, the brake and the gears of the used vehicles in order to do so. The longitudinal controller uses two loops; the outer loop computes a desired acceleration, which the inner loop uses as a reference. The outer loop uses acceleration, velocity and position information from the own vehicle and preceding vehicles. These three states first have to be estimated from various sensors in the vehicle. An Extended Kalman Filter (EKF) has been used for fusing the signals from the different sensors. The used signals are available from DGPS and from inertial sensors on the test-vehicles. Real-life experiments with the proposed algorithm for a cluster of three vehicles will demonstrate the usefulness of the approach.

Persistency of excitation in subspace predictive control

Abstract This paper presents a method that ensures persistency of excitation for subspace predictive control. This control method is characterized by the combination of a predictive control law with a subspace predictor. The subspace predictor is continuously being adapted to the controlled system by using input-output data from this system. For this purpose the input-output data should be persistently exciting. In this paper a method is proposed to ensure persistency of excitation by adding a term to the cost function used by the predictive control law. This term is designed such that only the least excited directions of the input space are additionally excited. An advantage of the method is that the optimization problem that needs to be solved for the predictive controller can still be solved by using quadratic programming. The proposed excitation method is evaluated in simulation on a detailed nonlinear model of a transport aircraft. The simulation results clearly show the usefulness of the proposed method.

Subspace Predictive Control Applied to Fault-Tolerant Control

Subspace identification is a technique that can be used for identification of state-space models from input-output data. This technique has drawn considerable interest in the last two decades [1, 2], especially for linear time-invariant systems. A reason for this is the efficient way in which models are identified for systems of high order and with multiple inputs and outputs. Subspace identification can be used to form a subspace predictor for prediction of future outputs from past input-output data and a future input-sequence. This subspace predictor can be computed without realization of the actual state-space models, which significantly reduces computational requirements. In [3] the subspace predictor has been combined with model predictive control [4], resulting in a control algorithm that has been given the name subspace predictive control (SPC). In SPC, the output predicted by the subspace predictor is part of the cost function of the predictive controller. As a result of the subspace predictor being generated completely from input-output data, the SPC algorithm is a data-driven one.

Experimental evaluation of a co-operative driving setup based on inter-vehicle communication.

Abstract This paper discusses field experiments conducted to evaluate a co-operative longitudinal controller for a cluster of vehicles with inter-vehicle communication (IVC). The controller can activelycontrol throttle, brake and gears of the used vehicles in order to realize a smooth traffic flow. For this purpose it uses two loops; the outer loop computes a desired acceleration, which the inner loop uses as a reference. The outer loop uses acceleration, velocity and position information from the own vehicle and preceding vehicles. This information is obtained by a DGPS-based Extended Kalman Filter (EKF). Real-life experiments have been performed on a closed test track. These experiments involved three vehicles of which the two rear vehicles had to anticipate automatically on the longitudinal manoeuvres of the front vehicle.

Model Weight Estimation for FDI Using Convex Fault Models

: A method is proposed for modeling a large number of faults in a system by a convex combination of a limited number of fault models that form a model set. The fault models in this model set, correspond to the maximum and minimum expected faults for faults that can occur partially. In this way, partial faults can be represented by a convex combination of the models from the model set. The identification of faults is performed by estimating the weights of the models from the model set. A set of linearized models of a Boeing 747 aircraft is used to display the effectiveness of the proposed method. This model set also includes models of the aircraft that correspond to faults that occurred during the disastrous crash of EL AL flight 1862 in 1992.