Abstracts
126
078 Traekin~ Using a Random Sampling Algorithm J.A. O'SnUivan, NLI. MIHer, A. Srivastava, D.L. Snyder, pp 361-364 A random sampling algorithm for nonlinear state estimation is formulated. Special cases of the algorithm are presented. A continuous Markov process is defined, along with a measurement distribution. From these, the posterior distribution on the states is derived, The sampling algorithm generates realizations from this distribution. A parallel implementation of the algorithm is presented. An application discussed is the problem of tracking radar targets using data from multiple sensors.
079 Trucking with Consistent Converted Measurements vs. the E K F D. Lerro, Y. Bar-Shalom, pp 365-368 In tracking applications, target motion is usually best modeled using Cartesian coordinates. However, in most systems the target position measurements are provided in terms of range and azimuth (bearing). An accurate means of tracking with debiased consistent converted measurements is presented, which accurately accounts for sensor inaccuracies over all practical geometries and accuracies. This method is compared to the mixed-coordinate EKF approach, as well as the standard converted measurement approach. The new approach is shown to be more accurate in terms of position and velocity errors, and provides consistent estimates for all practical simulations. The combination of parameters for which debiasing is needed is presented in explicit form.
080 Fault Detection in Idnear Discrete Dynamic Systems using Genendized-IAkelihood-Ratio Technique S. Tanaka, pp369-376 The generalized-likelihood-ratio (GLR) technique is used for fault detection in linear discrete dynamic systems. Based on fault detectability, three detection methods are fast introduced; a reduced order step-hypothesized GLR (SHGLR) method, a reduced-order tracking functional subapace method, which makes use of the system information about input and observation, and a pattern-recognition-based method, which recognizes the pattern of the curve of the reduced-order SHGLR to detect the fault. Finally, the system's robustness in the face of uncertainties is considered.
081 Robust Fault Detection Method Accounting for Moddlin£ Errors in Uncertain Systems Oh-Kyu Kwon, W.H. Kwon, J.H. Lee, pp 377-382 This paper deals with the fault-detection problem in uncertain linear/nonlinear systems having both undermedelling and noise. A robust fault-detection method is presented which accounts for the effects of the variance and bias errors caused by the noise and the model mismatch, respectively. The model mismatch includes linearization error as well as undermodelling. Comparisons are made with alternative fault-detection methods which do not account for model mismatch or linearization errors. The new method is shown to have good performance on a number of simulated systems.
082 An Approach to Residual Generator and Evaluator Design and Synthesis X. Dlng, P.M. Frank, L. G u t , pp 383-386 Problems related to fault detection for uncertain dynamic systems are studied. The objective of this study is to develop an approach to residual generator and evaluator design and synthesis. The basic idea of the new approach is to design residual generators and establish thresholds in such a way that the prescribed requirements for the size of detectable faults and the rate of false alarms are met.
isolation purposes. Next, an approach to the design of a decision logic suitable for the proposed residual generation scheme is described.
084 A General Failure Detection, Isolation and Accommodation System with Model Uncertainty and Measurement Noise Chia-Chi Tsui, pp 391-398 This paper establishes a nonzero threshold on a normally zerovalued signal, for the effect of model uncertainty and measurement noise in a general failure-detection, -isolation, and accommodation system. The corresponding detectable failure level is also established. Such a threshold was previously established only for a single failure detector, while the faihireisolation system of this paper requires a set of robust failure detectors. Based on this failure-isolation system, an adaptive state feedback control scheme is also proposed for failure accommodation. This adaptive state feedback system is very powerful and easily implementable, and can be systematically designed.
085 On Construction of Sequential Procedures for Detection of a Chonge-Point of Parameters in Stochastic Dynamic
Systems S.E. Vorobejchlkov, V.V. Konev, pp 399-402 The problem of detecting a stepwise parameter change in linear stochastic systems is considered. The system parameters before and after the disruption time are assumed known. New sequential procedures of disruption detection based on the cumulative sums method are proposed. The main innovation is that the statistics used in the algorithm are calculated by a set of observations of random volume. Also, quantization of the statistics in the cumulative sums algorithm enables the chief characteristics: the mean time between false alarms and the mean delay time of disruption detection, to be obtained analytically. The mean delay time increases logarithmically with the growth of mean time between false alarms.
086 Nineteen ML Estimators for Model Structure Selection F. Gustafsson, H. Hjalmarsson, pp 403-408 Classical approaches to determining a good model structure are based on hypothesis tests and information-based criteria. Recently, model structure has been considered as a stochastic variable, and standard estimation techniques have been proposed. However, a number of prior choices in the problem formulation are crucial for the estimators' behaviour. The contribution of this paper is to clarify the role of the prior choices, to examine all conceivable possibilities, and to show, in a linear regression framework, which estimators are consistent. For auto-regressive models, the prior assumption of stability is investigated, and the estimator for the model order and the parameters themselves is given.
087 Identiflahility in Blind Equalization F. Gustafsson, B. Waldberg, pp 409-412 This paper is concerned with estimation of the input sequence of an unknown linear system, given noisy measurements of the output signal. It is assumed that the system can be described by a finite impulse response model and that the input signal belongs to a finite alphabet. This problem formulation is motivated by digital communication systems, where it is called the "blind equalization" problem. The objective is to give conditions under which it is actually possible to solve this problem. For the noise-free case the blind equalization problem reduces to solving a non-linear equation system. Sufficient conditions to guarantee uniqueness of the solution are given.
083 Design of Redundancy Relations for Failure Detection and Isolation by Constrained Optimization M. Khmaert, pp 387-390
088 Simultaneous Solution of Smoothing, Filtering a n d Prediction Problems Using Integral Equations
The design of a fault-detection and -isolation (FI)I) system for linear time-invariant processes submitted to additive failures is considered, The paper fast concentrates on the problem of residual generation. The redundancy relations are obtained by minimizing a quadratic cost function under non-convex quadratic inequality constraints. The constraints express, in some way, the desired performance of the FDI system, for both detection and
The goal of this paper is to present an algorithm for simultaneous solution of stochastic smoothing, filtering and prediction problems. This algorithm is obtained when a stochastic process is described by a linear stochastic Ito-Volterra integral equation. The Kalman filtering problem formulation is used in the given paper. It has been shown that the gain of the optimal estimator and the covariauce of the estimation error are connected by a system of
D.G. Maksarov, pp 413-416