Mario Versaci

Microwave Devices and Antennas Modelling by Support Vector Regression Machines

Development of fast and accurate models of microwave devices and antennas is of paramount importance in computer-aided design and circuit analysis. At this purpose, artificial neural networks (ANNs) have been extensively exploited in technical literature. However, in the last years support vector machines (SVMs) developed by Vapnik are gaining popularity due to many attractive features capable to overcome the limitations connected to ANNs. In this work, support vector regression machines (SVRMs) modelling performances are investigated and compared with ANNs performances by means of several cases of study

Fuzzy neural identification and forecasting techniques to process experimental urban air pollution data

This paper focuses on the processing of experimentally measured pollution data. Measuring locally both air quality parameters and atmospheric data can show how complex can be their interrelations and how they change spatially. Furthermore, apart from physical and biochemical dependencies, two important aspects need to be incorporated in the model, traffic data and topographic information, like presence and configuration of buildings and roads. Since estimating the evolution of pollutant in the urban air can have significant economic impact already on a short term basis as well as relevant consequences on public health on a medium-long term scale, various interdisciplinary researches are under way on this subject. In this work, we pursue two goals. The first one is to derive a representative model of the multivariate relationships that should be able to reproduce local interactions; the second goal of the paper is to predict, when possible, the short term evolution of pollutants in order to prevent the onset of above threshold levels of pollutants that can be dangerous to humans. The threshold levels of interest are fixed by both EU recommendations and regional regulations. As a by-product of the research, we could derive some directives to be supplied to local authorities to properly organize car traffic in advance based on the estimated parameters. The case study here proposed is that of Villa San Giovanni, a small town at the tip of Italy, located just in front of Sicily, on the Messina Strait. This is a significant case, since the city is affected by the heavy traffic directed (and coming from) Sicily. The main results here reported include the short time prediction of the concentration of hydrocarbons (HC) in the local air, the comparison between different methods based on fuzzy neural systems, and the proposal of local models of non-linear interactions among traffic, atmospheric and pollution data. Additionally, comments on a longer horizon forecast are given.

Resonant frequency evaluation of microstrip antennas using a neural-fuzzy approach

The accurate evaluation of the resonant frequency of microstrip antennas is a key factor to guarantee their correct behavior. To this aim, the Method of Moments technique is currently employed. A fast technique to evaluate the resonant frequency of microstrip antennas using neuro-fuzzy networks, is proposed. Numerical results obtained by using this technique agree quite well with the Method of Moments results. The proposed technique can be fruitfully used in microwave CAD applications.

Fuzzy time series approach for disruption prediction in Tokamak reactors

Disruption in a Tokamak reactor is a sudden loss of confinement that can cause damage to the machine walls and support structures. In this paper, we propose the use of a fuzzy time series (FTS) approach for detection of disruption in Tokamaks. In particular, two-factors FTS models have been exploited for the prediction of the time to disruption in the ASDEX-Upgrade machine. The concept of fuzzy logic is used taking into account that previous techniques make use of expert knowledge for deciding about the onset of a disruption.

Clustering of entropy topography in epileptic electroencephalography

Epileptic seizures have been considered sudden and unpredictable events for centuries. A seizure seems to occur when a massive group of neurons in the cerebral cortex begins to discharge in a highly organized rhythmic pattern, then it develops according to some poorly described dynamics. As proved by the results reported by different research groups, seizures appear not completely random and unpredictable events. Thus, it is reasonable to wonder when, where and why the epileptogenic processes start up in the brain and how they result in a seizure. In order to detect these phenomena from the very beginning (hopefully minutes before the seizure itself), we introduced a technique, based on entropy topography, that studies the synchronization of the electric activity of neuronal sources in the brain. We tested it over 3 EEG data set from patients affected by partial epilepsy and 25 EEG recordings from patients affected by generalized seizures as well as over 40 recordings from healthy subjects. Entropy showed a very steady spatial distribution and appeared linked to the brain zone where seizures originated. A self-organizing map-based spatial clustering of entropy topography showed that the critical electrodes shared the same cluster long time before the seizure onset. The healthy subjects showed a more random behaviour.

Deep Learning Representation from Electroencephalography of Early-Stage Creutzfeldt-Jakob Disease and Features for Differentiation from Rapidly Progressive Dementia.

A novel technique of quantitative EEG for differentiating patients with early-stage Creutzfeldt-Jakob disease (CJD) from other forms of rapidly progressive dementia (RPD) is proposed. The discrimination is based on the extraction of suitable features from the time-frequency representation of the EEG signals through continuous wavelet transform (CWT). An average measure of complexity of the EEG signal obtained by permutation entropy (PE) is also included. The dimensionality of the feature space is reduced through a multilayer processing system based on the recently emerged deep learning (DL) concept. The DL processor includes a stacked auto-encoder, trained by unsupervised learning techniques, and a classifier whose parameters are determined in a supervised way by associating the known category labels to the reduced vector of high-level features generated by the previous processing blocks. The supervised learning step is carried out by using either support vector machines (SVM) or multilayer neural networks (MLP-NN). A subset of EEG from patients suffering from Alzheimers Disease (AD) and healthy controls (HC) is considered for differentiating CJD patients. When fine-tuning the parameters of the global processing system by a supervised learning procedure, the proposed system is able to achieve an average accuracy of 89%, an average sensitivity of 92%, and an average specificity of 89% in differentiating CJD from RPD. Similar results are obtained for CJD versus AD and CJD versus HC.

Fuzzy Characterization of Flawed Metallic Plates with Eddy Current Tests

Eddy Current Techniques (ECT) for Non-Destructive Testing and Evaluation (NDT/NDE) of conducting materials is one of the most application-oriented field of research within electromagnetism. In this work, a novel approach is proposed in order to characterize defects on metallic plates in terms of their depth and shape, starting from a set of experimental measurements. The problem is solved by means of a hybrid classification system based on Computation with Words (CWs) and Fuzzy Entropy (FE). They extract information about the specimen under test from the measurements. Main advantages of proposed approach are the introduction of CWs as well as the usage of the FE based minimization module, in order to improve flaw characterization by a low computational complexity system.

FEA Design and Misfit Minimization for In-Depth Flaw Characterization in Metallic Plates With Eddy Current Nondestructive Testing

Nondestructive testing techniques are more and more exploited in order to quickly and cheaply recognize flaws into the inspected materials. Within this framework, a concern of eddy current tests is the depth of penetration delta, above all in such applications as the control of steel beams. Thus, an optimal design of exciting coil is strictly required in order to reach as higher delta as possible. The aim of this paper is first to design, by finite-element analysis, and test by in-lab measurements, a suitable exciting coil. Subsequently, the inverse ill-posed problem for defect characterization, starting from experimental measurements, has been studied and regularized, in order to characterize the depth and the extension of defects.

A neuro-fuzzy approach to simulate the user mode choice behaviour in a travel decision framework

Abstract Increasing congestion on the main roads in urban areas pushes analysts to improve simulation of modal choice to obtain good estimation of the demand shares for different travel modes. In the literature, several kinds of behavioural models have been proposed in order both to evaluate the modal choice percentage in urban areas and to capture the travel behaviour by means of the estimation of some suitable parameters. However, behaviour is complex in itself and often standard models, even if sophisticated, cannot capture all the complex mechanisms underlying the user behaviour. In this paper, a neuro-fuzzy approach is proposed to extract the mode choice decision rules of travel users, by evaluating different sets of rules and different membership functions. Particularly, to determine which inputs are the most relevant in such decision process, fuzzy curves and surfaces have been taken into account. In this way, nonlinear effects can be considered.

A novel approach for detecting and classifying defects in metallic plates

In the field of nondestructive testing on defect identification in metallic plates, the shape reconstruction is still an open question. State-of-the-art technologies indeed enable the operator to locate the position of a defect but not its shape. The aim of this paper is to make a contribution to the solution of this side of the problem suggesting a novel methodology based on a neurofuzzy approach. Sugenos neurofuzzy inferences have been carried out for this purpose, as a first step in this direction. Fuzzy entropy was then exploited to measure how far is a given defect from a well-known depth. A sort of classification based on the depth of a defect has been performed this way.