Krzysztof Patan

Soft Computing Approaches to Fault Diagnosis for Dynamic Systems

Recent approaches to fault detection and isolation for dynamic systems using methods of integrating quantitative and qualitative model information, based upon soft computing (SC) methods are surveyed and studied in some detail. SC methods are considered an important extension to the quantitative model-based approach for residual generation in fault detection and isolation (FDI). When quantitative models are not readily available, a correctly trained neural network (NN) can be used as a non-linear dynamic model of the system. The paper describes some powerful NN methods, taking into account the dynamic as well as non-linear system behaviour. Sometimes, further insight is required as to the explicit behaviour of the model-involved and it is then that fuzzy and even neurofuzzy methods come to their own in data-driven FDI applications. The paper also discusses the use of evolutionary programming tools for observer and NN design. The paper provides many powerful examples of the use of SC methods for achieving good detection and isolation of faults in the presence of uncertain plant behaviour, together with their practical value for fault diagnosis of real process systems.

Stability Analysis and the Stabilization of a Class of Discrete-Time Dynamic Neural Networks

This paper deals with problems of stability and the stabilization of discrete-time neural networks. Neural structures under consideration belong to the class of the so-called locally recurrent globally feedforward networks. The single processing unit possesses dynamic behavior. It is realized by introducing into the neuron structure a linear dynamic system in the form of an infinite impulse response filter. In this way, a dynamic neural network is obtained. It is well known that the crucial problem with neural networks of the dynamic type is stability as well as stabilization in learning problems. The paper formulates stability conditions for the analyzed class of neural networks. Moreover, a stabilization problem is defined and solved as a constrained optimization task. In order to tackle this problem two methods are proposed. The first one is based on a gradient projection (GP) and the second one on a minimum distance projection (MDP). It is worth noting that these methods can be easily introduced into the existing learning algorithm as an additional step, and suitable convergence conditions can be developed for them. The efficiency and usefulness of the proposed approaches are justified by using a number of experiments including numerical complexity analysis, stabilization effectiveness, and the identification of an industrial process

Nonlinear model predictive control of a boiler unit

Nonlinear model predictive control of a boiler unit: A fault tolerant control study This paper deals with a nonlinear model predictive control designed for a boiler unit. The predictive controller is realized by means of a recurrent neural network which acts as a one-step ahead predictor. Then, based on the neural predictor, the control law is derived solving an optimization problem. Fault tolerant properties of the proposed control system are also investigated. A set of eight faulty scenarios is prepared to verify the quality of the fault tolerant control. Based of different faulty situations, a fault compensation problem is also investigated. As the automatic control system can hide faults from being observed, the control system is equipped with a fault detection block. The fault detection module designed using the one-step ahead predictor and constant thresholds informs the user about any abnormal behaviour of the system even in the cases when faults are quickly and reliably compensated by the predictive controller.

Optimal observation strategies for model-based fault detection in distributed systems

In model oriented diagnostics of real-world systems, the problems of structural identification and parameter estimation are of crucial importance. They require a properly designed schedule of measurements in such a way as to obtain possibly the most informative observational data. The aim of this work is to develop a novel approach to fault detection in distributed systems based on the maximization of the power of parametric hypothesis test, which verifies the nominal state of the considered system. The optimal locations of sensors are determined using the performance index operating on the Fisher Information Matrix. A general scheme is then proposed and tested on a computer example regarding an advection-diffusion problem.

Stochastic learning methods for dynamic neural networks: simulated and real-data comparisons

In the paper some stochastic methods for dynamic neural network training are presented and compared. The considered network is composed of dynamic neurons, which contain inner feedbacks. This network can be used as a part of fault diagnosis system to generate residuals. Classical optimisation techniques, based on back propagation idea, suffer from many well-known drawbacks. Two stochastic algorithms are tested as training algorithms to overcome these difficulties. Efficiency of proposed learning methods is checked on two examples: modelling of an unknown linear dynamic system basing on simulated data and modelling of the actuator behaviour in the first section of the evaporation station in the Sugar Factory, Lublin using real data measurements. In these two significant examples, the stochastic learning algorithms are extensively compared from many different perspectives.

Neural Network-Based Model Predictive Control: Fault Tolerance and Stability

This brief deals with nonlinear model predictive control designed for a tank unit. The predictive controller is realized by means of a recurrent neural network, which acts as a one-step ahead predictor. Then, based on the neural predictor, the control law is derived solving an optimization problem. An important issue in control theory is stability of the control system. In this brief, this problem is investigated by showing that a cost function is monotonically decreasing with respect to time. The derived stability conditions are then used to redefine a constrained optimization problem in order to calculate a control signal. As the automatic control system can prevent faults from being observed, the control system is equipped with a fault diagnosis block. It is realized by means of a multivalued diagnostic matrix, which is determined on the basis of residuals calculated using a set of partial models. Each partial model is designed in the form of a recurrent neural network. This brief proposes also a methodology of compensating sensor, actuator, as well as process faults. When a sensor fault is isolated, the system estimates its size and, based on this information, the controller is fed with a determined, close to real, tank level value. Actuator and process faults can be compensated due to application of an unmeasured disturbance model.

Network of dynamic neurons in fault detection systems

The neural-residual generator (NRG) construction for a dynamic system is the goal of this paper. The neural networks modeling the dynamic system are constructed with dynamics models of artificial neurons, which contain inner feedbacks. Such a model consists of an adder module, a linear dynamic system and a nonlinear activation function. Two optimization problems are solved with NRG construction: searching for an optimal network architecture and dynamic neuron model (DNM) network training. The boiler unit of a power plant is chosen as a modeled dynamic system.The cascade network of dynamic neurons (CNDN) as a neural-residual generator for fault detection in a dynamic systems is considered. The neural network is composed of dynamic neurons, which contain inner feedbacks. These neurons consists of an adder module, a linear dynamic system (IIR filter), and a non-linear activation function. The cascade-correlation algorithm is used for network architecture and parameter allocation. As a illustrative example of the diagnosed dynamic system, the two-tank system is chosen. The proposed approach is useful in neural modelling of dynamic system for FDI (Fault Detection and Isolation).

Approximation of state-space trajectories by locally recurrent globally feed-forward neural networks

The paper deals with investigating approximation abilities of a special class of discrete-time dynamic neural networks. The networks considered are called locally recurrent globally feed-forward, because they are designed with dynamic neuron models which contain inner feedbacks, but interconnections between neurons are strict feed-forward ones like in the well-known multi-layer perceptron. The paper presents analytical results showing that a locally recurrent network with two hidden layers is able to approximate a state-space trajectory produced by any Lipschitz continuous function with arbitrary accuracy. Moreover, based on these results, the network can be simplified and transformed into a more practical structure needed in real world applications.

Application of the state space neural network to the fault tolerant control system of the PLC-controlled laboratory stand

This paper deals with the design of a fault tolerant control system for a laboratory stand. With the application of a state space neural network it is possible to design both the nonlinear model and the observer of the considered plant. Analysing outputs of those models, it is possible to carry out fault detection. In order to cope with uncertainties of the model, a robust fault detection scheme is used which is based on the model error modelling technique. When a fault is detected, the fault tolerant control starts to compensate the fault effect. This is achieved through a proper recalculation of a control law. The new control law is obtained by adding an auxiliary signal to the standard control. This auxiliary control constitutes the additional control loop which can affect the stability of the entire control system. Therefore, stability of the proposed control scheme based on the Lyapunov direct method is also investigated. Finally, the approach is tested on the fluid flow and pressure control laboratory stand.

Model Predictive Control of the Two Rotor Aero-Dynamical System Using State Space Neural Networks with Delays

This paper deals with the application of state space neural network model with delays to design a model predictive control for a laboratory stand of the Two Rotor Aero-dynamical system. The work describes approach based on the so-called instantaneous linearisation of the already trained nonlinear state space model of the system. With obtained linear model it is possible to derive a vector of future controls based on the minimisation of the cost function within one optimisation window. Repeating procedure in each step of simulation and applying the obtained control signal allows for efficiently control of the nonlinear systems. All data used in experiments is obtain from the real-time laboratory stand which is working in Matlab/Simulink RTW environment.