Zeljko Djurovic

Detection of epileptiform activity in EEG signals based on time-frequency and non-linear analysis

We present a new technique for detection of epileptiform activity in EEG signals. After preprocessing of EEG signals we extract representative features in time, frequency and time-frequency domain as well as using non-linear analysis. The features are extracted in a few frequency sub-bands of clinical interest since these sub-bands showed much better discriminatory characteristics compared with the whole frequency band. Then we optimally reduce the dimension of feature space to two using scatter matrices. A decision about the presence of epileptiform activity in EEG signals is made by quadratic classifiers designed in the reduced two-dimensional feature space. The accuracy of the technique was tested on three sets of electroencephalographic (EEG) signals recorded at the University Hospital Bonn: surface EEG signals from healthy volunteers, intracranial EEG signals from the epilepsy patients during the seizure free interval from within the seizure focus and intracranial EEG signals of epileptic seizures also from within the seizure focus. An overall detection accuracy of 98.7% was achieved.

Classification of EEG signals for detection of epileptic seizures based on wavelets and statistical pattern recognition

The electroencephalogram (EEG) signal is very important in the diagnosis of epilepsy. Long-term EEG recordings of an epileptic patient contain a huge amount of EEG data. The detection of epileptic activity is, therefore, a very demanding process that requires a detailed analysis of the entire length of the EEG data, usually performed by an expert. This paper describes an automated classification of EEG signals for the detection of epileptic seizures using wavelet transform and statistical pattern recognition. The decision making process is comprised of three main stages: (a) feature extraction based on wavelet transform, (b) feature space dimension reduction using scatter matrices and (c) classification by quadratic classifiers. The proposed methodology was applied on EEG data sets that belong to three subject groups: (a) healthy subjects, (b) epileptic subjects during a seizure-free interval and (c) epileptic subjects during a seizure. An overall classification accuracy of 99% was achieved. The results confirmed that the proposed algorithm has a potential in the classification of EEG signals and detection of epileptic seizures, and could thus further improve the diagnosis of epilepsy.

Consensus based overlapping decentralized fault detection and isolation

In this paper a new distributed fault detection and isolation (FDI) methodology is proposed in the form of a multi-agent network representing a combination of a consensus based FDI observer for residual generation and a consensus based decision making strategy for change detection, applicable in real time. The proposed observer is based on overlapping system decomposition and a combination between the local optimal stochastic FDI observers and a dynamic consensus strategy. It is shown how the proposed algorithm can generate residuals which provide, under general conditions concerning local models and the network topology, high efficiency, scalability and robustness. The proposed decision making strategy provides solutions for two particular cases: a) local detection for non-overlapping parts of the identified subsystems; b) a consensus based strategy for FDI in the overlapping parts. One selected example illustrates the applicability of the proposed methodology in practice.

Simulation of air turbulence signal and its application

The air turbulence signal is one of the most important disturbances that act during the flight of the aircraft, and therefore, the simulation of atmospheric turbulence is of considerable importance in the evaluation of an aircraft performance in real conditions. Passing a white stochastic sequence through the adequately designed linear filter, the obtained signal at its output will be with a power spectral density very close to the von Karman power spectral density model of air turbulence. The paper presents an analysis of the influence of the filter order and an optimization procedure for filter parameters tuning, in order to obtain a suitable approximation of the von Karman spectra.

Fuzzy-Based Controller for Differential Drive Mobile Robot Obstacle Avoidance

Abstract This paper presents a new methodology for the avoidance of one or more obstacles, and for the navigation of a differential-drive mobile robot. The approach is based on fuzzy logic with virtual fuzzy magnets, and represents a reactive controller for navigation through an unknown environment to the ultimate target. Relative parameters of the obstacle in the robots way are determined at the preprocessing stage and the algorithm is therefore applicable to obstacles of different sizes. The algorithm, designed to avoid a single stationary obstacle, was generalized and successfully applied in a multiple-obstacle navigation scenario. The efficiency of the algorithm is illustrated by computer simulations using the kinematic model of a mobile robot.

QQ-plot based probability density function estimation

We present a new algorithm for the estimation of probability density functions (PDF). This founds a large number of applications in the context of statistical signal processing problems, such as detection, estimation, filtering or pattern recognition and classification. Our approach relies on the QQ-plot technique. The estimates of the first and second order statistics of the observed random data are used together with a suboptimal piecewise linear approximation of the QQ-plot, yielding a new class of PDF estimator. We describe the algorithm and test it in comparison with other techniques, showing that our approach provides better results.

Combustion distribution control using the extremum seeking algorithm

Quality regulation of the combustion process inside the furnace is the basis of high demands for increasing robustness, safety and efficiency of thermal power plants. The paper considers the possibility of spatial temperature distribution control inside the boiler, based on the correction of distribution of coal over the mills. Such control system ensures the maintenance of the flame focus away from the walls of the boiler, and thus preserves the equipment and reduces the possibility of ash slugging. At the same time, uniform heat dissipation over mills enhances the energy efficiency of the boiler, while reducing the pollution of the system. A constrained multivariable extremum seeking algorithm is proposed as a tool for combustion process optimization with the main objective of centralizing the flame in the furnace. Simulations are conducted on a model corresponding to the 350MW boiler of the Nikola Tesla Power Plant, in Obrenovac, Serbia.

Neural network ensemble for power transformers fault detection

Electrical transformers are the most important elements in the process of transmission and distribution of electricity. Depending on the size and position of the transformer, the sudden device failure can cause tremendous damage. Neural networks are widespread technique for transformer health monitoring. Neural Network Ensembles are an advanced neural technique that improves the accuracy and reliability in the transformers health diagnosis and failure prognosis. This paper describes a technique how to identify causal relation of dissolved gases in transformers oil and the current state of the transformers health. The described algorithm improves the interpretation of results obtained by dissolved gas analysis (DGA) technique. The most important result of this algorithm is a timely and reliable prediction of transformers failure based on incipient faults detection.

On signal-to-noise ratio estimation

A new simple algorithm for estimating signal-to-noise ratio (SNR) for a signal consisting of one sinusoid in white Gaussian noise is proposed in this paper. Algorithm is based on autocorrelation and modified covariance methods for AR (Autoregressive) spectral estimation. The validity of the algorithm is examined by comparing its SNR estimate with SNR estimate obtained by sinusoid magnitude estimation using Pisarenko harmonic decomposition method and noise variance estimation using modified covariance method. By a large number of simulations this algorithm has proven itself as a comparably precise even in case of significantly noise-contaminated sinusoidal signal.

The robustness of the differential quantizer in the case of the variable signal to noise ratio

In this paper a short theoretical overview of differential quantizer and its implementations is given. Afterward, the effect of the order of prediction in differential quantizer and the effect of the difference in order of predictor in the input and output of differential quantizer is analyzed. Then it was proceeded with the examination of the robustness of the differential quantizer in the case in which a noise signal is brought to the input of the differential quantizer, instead of the clean speech signal. The analysis was conducted with a uniform distribution, as well as the noise with the gaussian distribution, and the obtained results were adequately commented on. Also, experimentally a limit was set which refers to the intensity of the noise and still enable results which are better that a regular uniform quantizer. The whole analysis is done by using the fixed number of bits in quantization, i.e. 12-bit quantizer is used in all the implementations of differential quantizer. In the conclusion of this paper there is a discussion about the possibility of implementing a differential quantizer which will be able to recognize which noise attacks the system, and in addition to that, in what form it adapts its coefficients so that it at any moment acquires the optimal signal to noise ratio.