Marcel Luzar

Neural network-based robust actuator fault diagnosis for a non-linear multi-tank system

The paper is devoted to the problem of the robust actuator fault diagnosis of the dynamic non-linear systems. In the proposed method, it is assumed that the diagnosed system can be modelled by the recurrent neural network, which can be transformed into the linear parameter varying form. Such a system description allows developing the designing scheme of the robust unknown input observer within H∞ framework for a class of non-linear systems. The proposed approach is designed in such a way that a prescribed disturbance attenuation level is achieved with respect to the actuator fault estimation error, while guaranteeing the convergence of the observer. The application of the robust unknown input observer enables actuator fault estimation, which allows applying the developed approach to the fault tolerant control tasks.

Actuators and sensors fault diagnosis with dynamic, state-space neural networks

In this paper, the actuators and sensors fault detection and localization using a system model is considered. To obtain the system model, the neural network modeling is used. The artificial feedforward neural network with dynamic neurons in the state-space representation is proposed. To estimate the neural network parameters, the Adaptive Random Search algorithm with projection is used. To identify, which of actuators or sensors is faulty, the system input estimator is proposed. The input and output residuals being the difference between the system input and output and its estimates are used to detect and isolate the faults. The final part of the paper presents an application study, which clearly confirms the effectiveness of the proposed approach.

Neural-network based robust predictive fault-tolerant control for multi-tank system

The main contribution of the paper was to propose a robust predictive fault-tolerant control scheme for a class of non-linear discrete-time systems that can be described with the LPV models using neural networks. Indeed, the contribution can be divided into a few important points: extension of the efficient predictive control to the robust case with exogenous external disturbances acting on the system, development of robust fault estimation and compensation scheme, and an integration of the developed schemes within a unified robust predictive fault-tolerant control framework. The proposed approach was applied to the benchmark example of the multi-tank system. The achieved results show the performance of the high performance of the proposed approach. In spite of the incontestable appeal of the proposed approach there are still some points, which may further improve its effectiveness. Indeed, in the proposed approach it is assumed the the state is available and, hence a natural approach is to relax this assumption by the introduction of a suitable state estimation strategy.

A robust fault-tolerant model predictive control for linear parameter-varying systems

The paper deals with the problem of robust fault-tolerant model predictive control for non-linear discrete-time systems described by the Linear Parameter-Varying model. The proposed approach is based on a multi-stage stage procedure. Robust controller is designed without taking into account the input constraints related with the actuator saturation and deals with previously estimated faults. Thus, to check the compensation feasibility, employed robust invariant set takes into account the input constraints. If the current state does not belong to the robust invariant set, then a predictive control is performed in order to enhance the invariant set. This appealing phenomenon makes it possible to enlarge the domain of attraction, which makes the proposed approach an efficient solution for the fault-tolerant control. The final part of the paper shows an illustrative example for the variable-speed variable-pitch wind turbines.

Robust H ∞ actuator fault diagnosis with neural network

The paper deals with the problem of a robust actuator fault diagnosis for Linear Parameter-Varying (LPV) systems with Recurrent Neural-Network (RNN). The preliminary part of the paper describes the derivation of a discrete-time polytopic LPV model with RNN. Subsequently, a robust fault detection, isolation and identification scheme is developed, which is based on the observer and H∞ framework for a class of nonlinear systems. The proposed approach is designed in such a way that a prescribed disturbance attenuation level is achieved with respect to the actuator fault estimation error while guaranteeing the convergence of the observer.

Neural-Network Based Robust FTC: Application to Wind Turbines

The paper deals with the problem of a robust fault diagnosis of a wind turbine. The preliminary part of the paper describes the Linear Parameter-Varying model derivation with a Recurrent Neural Network. The subsequent part of the paper describes a robust fault detection, isolation and identification scheme, which is based on the observer and \(\mathcal{H}_{\infty}\) framework for a class of non-linear systems. The proposed approach is designed in such a way that a prescribed disturbance attenuation level is achieved with respect to the actuator fault estimation error while guaranteeing the convergence of the observer. Moreover, the controller parameters selection method of the considered system is presented. Final part of the paper shows the experimental results regarding wind turbines, which confirms the effectiveness of proposed approach.

Robust MPC for a non-linear system – a neural network approach

The aim of the paper is to design a robust actuator fault-tolerant control for a non-linear discrete-time system. Considered system is described by the Linear Parameter-Varying (LPV) model obtained with recurrent neural network. The proposed solution starts with a discretetime quasi-LPV system identification using artificial neural network. Subsequently, the robust controller is proposed, which does not take into account actuator saturation level and deals with the previously estimated faults. To check if the compensation problem is feasible, the robust invariant set is employed, which takes into account actuator saturation level. When the current state does not belong to the set, then a predictive control is performed in order to make such set larger. This makes it possible to increase the domain of attraction, which makes the proposed methodology an efficient solution for the fault-tolerant control. The last part of the paper presents an experimental results regarding wind turbines.

Robust multi-model fault detection and isolation with a state-space neural network

This paper presents an design of a Robust Fault Detection and Isolation (FDI) diagnostic system by the means of state-space neural network. First, an solution utilizing multimodel technique is described, in which a Single-Input MultiOutput (SIMO) system is decomposed into a number of Multi-Input Single-Output (MISO) and Single-Input Single-Output (SISO) models. Application of such models makes possible to calculate a set of residual signals required in evaluation process with a Model Error Modelling (MEM) to obtain diagnostic signals. In turn, to isolate faults the diagnostic signals together with defined binary diagnostic table are applied. For experimental verification of the proposed approach, the laboratory stand of Modular Servo is chosen. All necessary data were gathered with the Matlab/Simulink software.

LPV system modeling with SSNN toolbox

The main objective of this paper is to develop and design a State Space Neural Network toolbox for a non-linear system modeling with an artificial state-space neural networks, which can be used in a model-based robust fault diagnosis and control. Such a toolbox is implemented in the MATLAB environment and it uses some of its predefined functions. It is designed in the way that any non-linear multi-input multi-output system is modeled and represented in a classical state-space form. The novelty of the proposed approach is that the final result of the identification process is the state, input and output matrices, not only the neural network parameters. The toolbox is equipped with the graphical user interface, which makes it useful for the users not familiar with a neural networks theory. This paper can be perceived as an extention of [1] on LPV systems.

SSNN toolbox for non-linear system identification

The aim of this paper is to develop and design a State Space Neural Network toolbox for a non-linear system identification with an artificial state-space neural networks, which can be used in a model-based robust fault diagnosis and control. Such toolbox is implemented in the MATLAB environment and it uses some of its predefined functions. It is designed in the way that any non-linear multi-input multi-output system is identified and represented in the classical state-space form. The novelty of the proposed approach is that the final result of the identification process is the state, input and output matrices, not only the neural network parameters. Moreover, the toolbox is equipped with the graphical user interface, which makes it useful for the users not familiar with the neural networks theory.