Showing papers by "Paul M. Frank published in 1997"
TL;DR: Some schemes extending the well-known diagnosis methods for linear systems to the nonlinear case are considered and the robustness of these schemes in presence of unknown inputs is discussed.
486 citations
TL;DR: The basic idea of a novel observer concept, the so-called “knowledge observer”, is introduced and the neural-network approach for residual generation and evaluation is outlined as well.
144 citations
TL;DR: A novel observer concept, the so-called knowledge observer, is introduced and an artificial neural network approach for residual generation and evaluation is outlined as well.
118 citations
TL;DR: The link between the bound of the modelling errors and the dynamic of the observer is given and the stability of the observers where the non-linearities are bounded is considered.
31 citations
01 Jun 1997
TL;DR: In this paper a qualitative observer is proposed, which is based on qualitative simulation and applied using a novel technique: observation filtering, which provides a framework in which the measured information can be utilized to reduce accumulated ambiguity and to avoid spurious solutions.
Abstract: Qualitative simulation is one of the main techniques in qualitative reasoning. It has the great potential to solve engineering problems, if only the difficulty of ambiguity can be overcome. On the other hand, in many applications measurements are often available. In this paper a qualitative observer is proposed, which is based on qualitative simulation and applied using a novel technique: observation filtering. This provides a framework in which the measured information can be utilized to reduce accumulated ambiguity and to avoid spurious solutions. Simulated results of qualitative simulation and qualitative observer are compared to show the effect of the new approach. As an application to fault detection and isolation, qualitative extensions of the dedicated observer scheme and the generalized observer scheme are presented.
18 citations
TL;DR: In order to select from the given residual data the important fault information a human support tool for the generation of a knowledge base for fault diagnosis will be presented in this paper.
14 citations
TL;DR: A fault diagnosis scheme is proposed which is based on the parameter estimation with neural networks approach, where the deviation of the estimated parameters to their nominal values are used for fault detection, isolation and identification of the fault severity.
10 citations
04 Jun 1997
TL;DR: In order to select from the given residual data the important fault information a human support tool for the generation of a knowledge base for fault diagnosis is presented in the paper.
Abstract: The goal of fault detection and isolation (FDI) is to decide whether and where a fault in the system under consideration has occurred avoiding wrong decisions that cause false alarms. To achieve a fault detection scheme which is robust in the sense of false alarms a combined quantitative/qualitative supervision system is used to detect and isolate faults. The quantitative part is used to generate fault symptoms (residuals) using a quantitative (mathematical) model of the process. These residuals contain the information about whether a fault has occurred or not. The next step in the FDI process is the residual evaluation. There exists a number of different residual evaluation techniques, for example simple threshold logic tests, statistical decision making, pattern recognition and decision making based on fuzzy logic or neural networks. The fundamental difficulty with residual evaluation is that residuals are normally uncertain, corrupted by noise, disturbances and, if the residuals are generated by model-based techniques, by modelling uncertainties. In order to select from the given residual data the important fault information a human support tool for the generation of a knowledge base for fault diagnosis is presented in the paper.
6 citations
TL;DR: In this paper, Markov chains are used to represent qualitative behaviors generated through qualitative simulation, and the filtering techniques of qualitative simulation are discussed under this setting, based on the qualitative simulation and the observation filtering technique, the qualitative observer is constructed.
4 citations
TL;DR: In this paper, practical aspects of non-linear black-box modelling applied to hydrologic processes are investigated, and two methods for modelling nonlinear dynamic systems are presented; the first model consists in a set of local linear sub-models, and the second one is based on qualitative relations between the process variables Both the proposed methods are used to identify a rainfall-runoff relationship on a real urban watershed.
3 citations
TL;DR: Two observer based methods, using fuzzy and analytical techniques, are investigated and compared for the residual generation of fault diagnosis procedure applied to a roller for plastic band based on fuzzy clustering.
04 Jun 1997
TL;DR: In this article, an explicit form of the algorithm using the multidimensional Taylor series expansion of the Kushner-Stratonovitch equations is given, no restriction on the upper bound of the order is needed.
Abstract: We present a sequential and workable approach for high order filtering of nonlinear dynamical systems. First, an explicit form of the algorithm using the multidimensional Taylor series expansion of the Kushner-Stratonovitch equations is given, no restriction on the upper bound of the order is needed. Next, we generalize the formula for the expectation of the product of four to Gaussian random variables in order to establish the final form of the filtering algorithm.
TL;DR: The identification of nonlinear dynamic plants using a fully connected Recurrent Neural Network (RNN) is discussed and new methods will be shown that allow an effective training of this kind of artificial neural networks for system modelling tasks.
01 Jan 1997
TL;DR: In this article, an online supervision concept based on analytical redundancy is introduced, based on mathematical models of physical systems and measurements taken to control these systems, a fault detection concept is developed which can detect actuator, component and sensor faults.
Abstract: In this chapter an online supervision concept based on analytical redundancy is introduced. Based on mathematical models of physical systems and measurements taken to control these systems, a fault detection concept is developed which can detect actuator, component and sensor faults. With this algorithm, signals are generated which allow a reliable decision as to whether a fault has occurred or not and in some cases even give information about the nature of the fault. As a practical example the application of this concept to a three tank system is described and results are shown.
TL;DR: In this article, an approach to the detection and isolation of faults in linear time invariant systems is presented based on state and parameter estimation two residual generators are defined, one is an observer-based residual which is used to detect faults and it is given in a parametrised form.
TL;DR: In this paper, the robust estimation of parameters and states for non-linear discrete-time systems is studied and the convergence mechanisms for a global and decentralized recursive algorithm are analyzed based on the Lyapunov approach taking linearisation errors into account.
01 Jul 1997
TL;DR: In this paper, an extension of the Friedland's separate bias estimation algorithm for linear systems with colored noise was proposed, where the property of the bias is nonlinear, random and time-varying with an unknown changing law.
Abstract: This paper gives an extension of the Friedland's separate-bias estimation algorithm of linear systems to a class of nonlineartime-varying stochastic systems with colored noise. The property of the bias maybe nonlinear, random and time-varying with an unknown changing law. The results of computer simulation demonstrate the effectiveness of the proposed algorithm.
TL;DR: In this paper, the generalized likelihood ratio (GLR) fault detector for linear systems is extended to nonlinear stochastic systems in closed-loops, where two state estimation schemes are adopted to provide the residual error series, one is the famous EKF, another one is a strong tracking filter(STF) and a continuous stirred tank reactor(CSTR).