Khashayar Khorasani

Facial expression recognition using constructive feedforward neural networks

A new technique for facial expression recognition is proposed, which uses the two-dimensional (2D) discrete cosine transform (DCT) over the entire face image as a feature detector and a constructive one-hidden-layer feedforward neural network as a facial expression classifier. An input-side pruning technique, proposed previously by the authors, is also incorporated into the constructive learning process to reduce the network size without sacrificing the performance of the resulting network. The proposed technique is applied to a database consisting of images of 60 men, each having five facial expression images (neutral, smile, anger, sadness, and surprise). Images of 40 men are used for network training, and the remaining images of 20 men are used for generalization and testing. Confusion matrices calculated in both network training and generalization for four facial expressions (smile, anger, sadness, and surprise) are used to evaluate the performance of the trained network. It is demonstrated that the best recognition rates are 100% and 93.75% (without rejection), for the training and generalizing images, respectively. Furthermore, the input-side weights of the constructed network are reduced by approximately 30% using our pruning method. In comparison with the fixed structure back propagation-based recognition methods in the literature, the proposed technique constructs one-hidden-layer feedforward neural network with fewer number of hidden units and weights, while simultaneously provide improved generalization and recognition performance capabilities.

Actuator Fault Detection and Isolation for a Network of Unmanned Vehicles

This technical note investigates development, design and analysis of actuator fault detection and isolation (FDI) filters for a network of unmanned vehicles. It is shown that actuator fault signatures in a network of unmanned vehicles are dependent and the network can be considered as an over-actuated system. An isolability index mu is defined for a family of fault signatures and a new structured residual set is developed that is selectively capable of properly detecting and isolating mu multiple faults in linear systems with dependent fault signatures, such as over-actuated systems. Our proposed algorithm is then applied to the actuator FDI problem in a network of unmanned vehicles configured according to centralized, decentralized and semi-decentralized architectures. A comparative analysis in terms of the capabilities and limitations of these architectures is performed. Simulation results presented for the formation flight of multiple satellites demonstrate the effectiveness of our proposed FDI algorithm.

A Recurrent Neural-Network-Based Sensor and Actuator Fault Detection and Isolation for Nonlinear Systems With Application to the Satellite's Attitude Control Subsystem

This paper presents a robust fault detection and isolation (FDI) scheme for a general class of nonlinear systems using a neural-network-based observer strategy. Both actuator and sensor faults are considered. The nonlinear system considered is subject to both state and sensor uncertainties and disturbances. Two recurrent neural networks are employed to identify general unknown actuator and sensor faults, respectively. The neural network weights are updated according to a modified backpropagation scheme. Unlike many previous methods developed in the literature, our proposed FDI scheme does not rely on availability of full state measurements. The stability of the overall FDI scheme in presence of unknown sensor and actuator faults as well as plant and sensor noise and uncertainties is shown by using the Lyapunovs direct method. The stability analysis developed requires no restrictive assumptions on the system and/or the FDI algorithm. Magnetorquer-type actuators and magnetometer-type sensors that are commonly employed in the attitude control subsystem (ACS) of low-Earth orbit (LEO) satellites for attitude determination and control are considered in our case studies. The effectiveness and capabilities of our proposed fault diagnosis strategy are demonstrated and validated through extensive simulation studies.

Adaptive control of flexible joint robots

The problem of designing a robust adaptive control strategy for flexible joint robot manipulators is considered. By utilising the concept of integral manifolds, reduced-order models of the flexible system are obtained. This makes it possible to develop adaptive control schemes for the flexible full-order system at the reduced-order level, which otherwise would be difficult due to ill-conditioning and the curse of dimensionality. Two common adaptive control strategies are examined: (i) the adaptive inverse dynamics and (ii) the Slotine and Li algorithm. It is shown how these standard rigid adaptive control techniques can be generalized and improved for successful application to flexible joint manipulators. The result is robust adaptive control laws which take both parametric and dynamic uncertainties into account. Numerical simulations for a two-link flexible joint manipulator illustrate the potential and advantages of the new adaptive schemes as compare to the two standard algorithms in the literature.<<ETX>>

New training strategies for constructive neural networks with application to regression problems

Regression problem is an important application area for neural networks (NNs). Among a large number of existing NN architectures, the feedforward NN (FNN) paradigm is one of the most widely used structures. Although one-hidden-layer feedforward neural networks (OHL-FNNs) have simple structures, they possess interesting representational and learning capabilities. In this paper, we are interested particularly in incremental constructive training of OHL-FNNs. In the proposed incremental constructive training schemes for an OHL-FNN, input-side training and output-side training may be separated in order to reduce the training time. A new technique is proposed to scale the error signal during the constructive learning process to improve the input-side training efficiency and to obtain better generalization performance. Two pruning methods for removing the input-side redundant connections have also been applied. Numerical simulations demonstrate the potential and advantages of the proposed strategies when compared to other existing techniques in the literature.

Fault identification in an AC-DC transmission system using neural networks

The authors explore the possibility of using neural networks to identify faults that can occur in an AC-DC power system. Three types of neural network models have been studied and are compared. These networks can sense AC bus voltages either as root mean square (RMS) values (with or without phase angle information) or as sampled instantaneous values of sine waves. Depending on which method is used, some confusion can occur in distinguishing a line to line fault from a remote AC fault. A delay of 1-2 cycles in detection of faults when using RMS values is expected due to the algorithm required for determining the RMS value. This may not be too critical in practice. However, where this delay is unacceptable, instantaneous values may be used. Based on the ability of these networks to distinguish reliably between different types of faults, appropriate control measures can be taken to improve the dynamic performance of the AC-DC power system. >

Application of a radial basis function (RBF) neural network for fault diagnosis in a HVDC system

The application of a radial basis function (RBF) neural network (NN) for fault diagnosis in an HVDC power system is presented in this paper. To provide a reliable pre-processed input to the RBF NN, a new pre-classifier is proposed. This pre-classifier consists of an adaptive filter (to track the proportional values of the fundamental and average components of the sensed system variables), and a signal conditioner which uses an expert knowledge base (KB) to aid the pre-classification of the signal. The proposed method of fault diagnosis is evaluated using simulations performed with the EMTP package.

An integral manifold approach for tip-position tracking of flexible multi-link manipulators

In this paper, a nonlinear control strategy for tip position trajectory tracking of a class of structurally flexible multilink manipulators is developed. Using the concept of integral manifolds and singular perturbation theory, the full-order flexible system is decomposed into corrected slow and fast subsystems. The tip-position vector is similarly partitioned into corrected slow and fast outputs. To ensure an asymptotic tracking capability, the corrected slow subsystem is augmented by a dynamical controller in such a way that the resulting closed-loop zero dynamics are linear and asymptotically stable. The tracking problem is then redefined as tracking the slow output and stabilizing the corrected fast subsystem by using dynamic output feedback. Consequently, it is possible to show that the tip position tracking errors converge to a residual set of O(/spl epsiv//sup 2/), where /spl epsiv/ is the singular perturbation parameter. A major advantage of the proposed strategy is that the only measurements required are the tip positions, joint positions, and joint velocities. Experimental results for a single-link arm are also presented and compared with the case when the slow control is designed based on the rigid-body model of the manipulator.

Satellite fault diagnosis using a bank of interacting Kalman filters

The main objective of this work is development and testing of a detection, isolation, and diagnosis algorithm based on interacting multiple model (IMM) filters for both partial (soft) and total (hard) reaction wheels faults in a spacecraft. This is shown to be accomplished under a number of different faulty mode scenarios for these actuators associated with the attitude control system (ACS) of a satellite. Various operating and faulty conditions due to changes and anomalies in the temperature, the power supply line voltage, and the loss of effectiveness of the torque and the current are considered in each reaction wheel associated with the three axes of the satellite. Once a fault mode is detected and isolated the recovery procedure can subsequently be engaged by invoking appropriate switching control strategies for the ACS. The application of a bank of interacting multiple Kalman filters for detection and diagnosis of anticipated reaction wheel failures in the ACS is described and developed. Compared with other model-based fault detection, diagnosis and isolation(FDDI) strategies developed in the control systems literature, our FDDI strategy is shown, through extensive numerical simulations, to be more accurate and robust with potential for extension to a number of other application areas.

End-point control of a flexible-link manipulator: theory and experiments

This paper focuses on the tip-position control of a single flexible link which rotates in the horizontal plane. The dynamic model is derived using a Lagrangian assumed modes method based on Euler-Bernoulli beam theory. The model is then linearized about an operating point. An output feedback control strategy that uses the principle of transmission zero assignment achieves tracking for this nonminimum phase linear time-invariant system. The control strategy consists essentially of two parts. The first part is an inner (stabilizing) control. Loop that incorporates a feedthrough term to assign the systems transmission zeros at desired locations in the complex plane, and a feedback term to move the systems poles to appropriate positions in the left-half plane. The second part is a feedback servo loop that allows tracking of the desired trajectory. The controller is implemented on an experimental test-bed.