Rastko R. Selmic

Neuro-Fuzzy Control of Industrial Systems with Actuator Nonlinearities

From the Publisher: About the Author Frank L. Lewis is Associate Director for Research and Head of the Advanced Controls, Sensors, and MEMS Group at the Automation and Robotics Research Institute at the University of Texas at Arlington. Javier Campos is Linux Consultant Engineer at Montavista Software, Inc. in Irving, Texas. Rastko Selmic is DSP Systems Engineer at Signalogic, Inc. in Dallas, Texas.

Wireless Sensor Network Modeling Using Modified Recurrent Neural Networks: Application to Fault Detection

This paper presents a dynamic model of wireless sensor networks (WSNs) and its application to a sensor node fault detection. Recurrent neural networks (RNNs) are used to model a sensor node, its dynamics, and interconnections with other sensor network nodes. The modeling approach is used for sensor node identification and fault detection. The input to the neural network is chosen to include delayed output samples of the modeling sensor node and the current and previous output samples of neighboring sensors. The model is based on a new structure of backpropagation-type neural network. The input to the neural network and topology of the network are based on a general nonlinear dynamic sensor model. A simulation example has demonstrated effectiveness of the proposed scheme.

Neural net backlash compensation with Hebbian tuning using dynamic inversion

A dynamic inversion compensation scheme is presented for backlash. The compensator uses the backstepping technique with neural networks (NN) for inverting the backlash nonlinearity in the feedforward path. Instead of a derivative, which cannot be implemented, a filtered derivative is used. Full rigorous stability proofs are given using filtered derivative. Compared with adaptive backstepping control schemes, we do not require the unknown parameters to be linear parametrizable. No regression matrices are needed. The technique provides a general procedure for using NN to determine the dynamic preinverse of an invertible dynamical system. A modified Hebbian algorithm is presented for NN tuning which yields a stable closed-loop system. Using this method yields a relatively simple adaptation structure and offers computational advantages over gradient descent based algorithms.

Adaptive fuzzy logic compensation of actuator deadzones

A deadzone compensator is designed for nonlinear systems using a fuzzy logic (FL) controller. The classification property of FL systems makes them a natural candidate for the rejection of errors induced by the deadzone, which has regions in which it behaves differently. A tuning algorithm is given for the FL parameters so that the deadzone compensation scheme becomes adaptive, guaranteeing small tracking errors and bounded parameter estimates. A rigorous proof of stability and performance is given and a simulation example verifies excellent performance. It is seen that the extra signal produced by the adaptive FL deadzone compensator can be considered as an adaptive dithering term. ©1997 by John Wiley & Sons, Inc.

Multimodel neural networks identification and failure detection of nonlinear systems

Multimodel identification and failure detection using neural networks (NN) is presented. It is an extension and application of nonlinear system identification using radial basis function NN. The state estimation error is proven to converge to zero asymptotically. Parameters of the identifier converge to the ideal parameters provided that persistency of excitation condition is fulfilled. Multiple model identification structure is analyzed, and its application to the multimodel failure detection is considered. Two simulation examples for NN identifiers are given. Simulation for intelligent multimodel failure detection using multi-neural networks identifiers is presented.

Decentralized detection and patching of coverage holes in wireless sensor networks

Detection and patching of coverage holes in Wireless Sensor Networks (WSNs) are important measures of Quality of Service (QoS) for security and other applications that emphasize sensor network coverage. In this paper, we model a WSN using simplicial complexes based on its communication graph by which the network can be represented as connections of sensor nodes without knowing exact locations of nodes. Thus, the coverage problem is converted to a connectivity problem under some assumptions presented in the paper. We discuss two major topics in this paper, namely sensor network coverage hole detection and patching. We present a novel, decentralized, coordinate-free, node-based coverage hole detection algorithm. The algorithm can be implemented on a single node with connectivity information gathered from one-hop away neighbors. Thus, the coverage hole detection algorithm can be run on individual nodes and does not require time-consuming, centralized data processing. The hole-patching algorithm is based on the concept of perpendicular bisector line. Every hole-boundary edge has a corresponding perpendicular bisector and new sensor nodes are deployed on hole-boundary bisectors. Deployment of new sensor nodes maintains network connectivity, while reduces coverage holes.

Detecting coverage holes in wireless sensor networks

Wireless sensor network coverage completeness is an important Quality of Service measure. It is frequently assumed that events occurring in the sensor field will always be detected. However, this is not necessary the case, particularly if there are holes in the sensor network coverage. This paper introduces a novel method for detection and relative localization of sensor network coverage holes in coordinate-free networks assuming availability of a network communication graph. We identify sensor nodes that bound coverage holes, called “hole boundary nodes”, by processing information embedded in a communication graph, which is non-planar in general. We create a hole-equivalent planar graph preserving a number and position of holes. Finally, we build a planar simplicial complex, called maximal simplicial complex, which contains the information regarding coverage holes. The proposed method is applicable for both coordinate-available and coordinate-free networks. Two implementation strategies for hole detection are provided, and they are each analyzed to compare runtime and accuracy. Simulation results show effectiveness of the hole detection algorithms.

Backlash compensation in discrete time nonlinear systems using dynamic inversion by neural networks

A dynamics inversion compensation scheme is designed for control of nonlinear discrete-time systems with input backlash. The compensator uses the backstepping technique with neural networks (NN) for inverting the backlash nonlinearity in the feedforward path. The technique provides a general procedure for using NN to determine the dynamics pre-inverse of an invertible discrete time dynamical system. A discrete-time tuning algorithm is given for the NN weights so that the backlash compensation scheme becomes adaptive, guaranteeing bounded tracking and backlash errors, and also bounded parameter estimates. A rigorous proof of stability and performance is given and a simulation example verifies the performance. Unlike standard discrete-time adaptive control techniques, no certainty equivalence assumption is needed.

Adaptive Neural Network Output Feedback Control of Nonlinear Systems with Actuator Saturation

An indirect adaptive neural network (NN) saturation compensator is presented for a class of nonlinear systems. Output feedback control is considered where only the system output is assumed to be measurable. The imposed actuator saturation is assumed to be unknown and treated as the system input disturbance. A NN-based state observer estimates derivatives of the output and another NN-based feedback controller is inserted into a feedforward path to capture the nonlinearities of the observed system and to cancel the effects of the unknown disturbances and the unknown saturation nonlinearity. The unknown system states identified by the NN observer are inputs of the NN-based controller. Two NNs interact together to achieve the desired performance. Both adaptive, neural network control laws and on line neural net weights tuning rules are rigorously derived based on feedback linearization and Lyapunov approach. The overall robust adaptive scheme guarantees that the states estimation errors, NN weights estimation errors, and output tracking errors are uniformly ultimately bounded. The simulation conducted indicates the proposed scheme can effectively estimate the unknown nonlinear system states and accommodate the unknown actuator constraints.

Backlash compensation in nonlinear systems using dynamic inversion by neural networks

A dynamic inversion compensation scheme is presented for backlash. The compensator uses the backstepping technique with neural networks (NN) for inverting the backlash nonlinearity in the feedforward path. The technique provides a general procedure for using NN to determine the dynamic pre-inverse of an invertible dynamical system. A tuning algorithm is presented for the NN backlash compensator which yields a stable closed-loop system.