F. M. Frattale Mascioli

Partial discharge pattern recognition by neuro-fuzzy networks in heat-shrinkable joints and terminations of XLPE insulated distribution cables

An identification technique is described, based on a developed adaptive fuzzy logic network, that enables the recognition of partial discharges (PD) generated by different defects in heat-shrinkable joints and terminations of XLPE insulated power distribution cables. It is shown that different sources of PD can be identified on the basis of fuzzy rules applied to a selection of parameters derived from PD-pulse phase and amplitude distributions. A comparison with other PD pattern recognition techniques based on traditional neural networks is presented and discussed.

Constructive algorithm for neuro-fuzzy networks

A constructive algorithm is proposed by merging the min-max and the ANFIS models in order to obtain neuro-fuzzy networks. The min-max model is used to determine an optimal set of IF-THEN rules by following a constructive procedure. By means of this set, the architecture of an ANFIS-like net is derived with good performances in terms of structural complexity, generalization capability and speed of convergence. Simulations are described to show the behavior of the proposed algorithm.

Adaptive resolution min-max classifier

This paper presents a new neuro-fuzzy classifier, inspired by the Simpsons (1992, 1993) min-max model. By relying on a constructive approach, it overcomes some undesired properties of the original min-max algorithm. In particular, training result does not depend on pattern presentation order and hyperbox expansion is not limited by a fixed maximum size, so that it is possible to have different covering resolutions. Consequently, the new algorithm yields less complex networks, thus increasing the generalization capability in accordance with learning theory paradigms. Several tests are presented for illustration.

ANFIS synthesis by hyperplane clustering

Adaptive neuro-fuzzy inference systems (ANFIS) are one of the most popular types of fuzzy neural networks. An usual approach to the synthesis of ANFIS networks is based on clustering a training set of numerical examples of the unknown mapping to be approximated. Several different clustering procedures can be adopted for this purpose, but most of them are affected by serious drawbacks. We propose a novel clustering approach in order to overcome these problems. It determines directly the consequent part of ANFIS rules; successively, the fuzzy antecedent part of each rule is determined by using a Min-Max classifier. The resulting ANFIS architecture is optimized by means of a constructive procedure, which we further propose in the paper. It allows us to determine automatically the optimal number of rules by applying well-known results of learning theory. Simulation tests and comparison with other techniques are discussed in order to prove the validity of the proposed approach.

Automatic training of ANFIS networks

In the present paper an automatic training procedure for adaptive neuro-fuzzy inference system (ANFIS) networks is presented. The initialization of the net is carried out by the /spl beta/-min-max fuzzy clustering procedure, which is a modified version of the original min-max technique by Simpson (1993). Parameter /spl beta/ affects the number, position and size of resulting clusters. Since different P values yield different initializations, the optimal one is chosen by applying a well known result of the learning theory, which states that, under the same condition of performance on training set, the net that shows the best generalization capability is the one which is characterized by the lowest structural complexity. An automatic backpropagation-like procedure is finally used to perform a fine tuning of the optimal net. Simulation tests and comparison with other non-automatic learning procedures are discussed.

A constructive EM approach to density estimation for learning

Density estimation based on a mixture of Gaussian components is particularly suited to the solution of function approximation problems. When dealing with numerical examples of the function to be approximated, the corresponding neural network architecture can be trained by using a clustering procedure based on the well-known EM algorithm. However, the latter is characterized by some serious drawbacks that we overcome in this paper. For we propose a constructive procedure that increases progressively the number of Gaussian components; it yields improvements of both the speed and the quality of the EM convergence. Moreover, it also drastically reduces the computational cost of the optimization procedure that we further propose in order to select automatically the optimal number of Gaussian components of the neural network. The performance of the proposed approach is compared in the paper with respect to well-known neural network approaches.

A pyramidal delayed perceptron

It is pointed out that choosing the number of layers and the number of neurons per layer is critical in the actual application of the multilayer perceptron. A method for overcoming this inconvenience is proposed. The method requires modifying the structure of the perceptron by introducing suitable delays between adjacent layers and by connecting all the neurons to be input. With this approach, it is possible to automatically yield both the configuration of the network and the values of the weights of the connections. The new type of perceptron has a pyramidal shape just like the digital perceptron introduced by J.J. Vidal (1988). >

A real time classifier for emotion and stress recognition in a vehicle driver

Recently there is a great interest in artificial systems able to understand and recognize human emotions. In this paper an Emotion Recognition System based on classical neural networks and neuro-fuzzy classifiers is proposed. Emotion recognition is performed in real time starting from a video stream acquired by a common webcam monitoring the users face. Neurofuzzy classifiers, in comparison with Multi Layer Perceptron trained by EBP algorithm, show very short training times, allowing applications with easy and automated set up procedures, to be used in a wide range of applications, from entertainment to safety. The algorithm yields very interesting performances and can be adopted to recognize emotions as well as possible pathological conditions of the individual to be monitored.

Automotive application of lithium-ion batteries: A new generation of electrode materials

New researches suggest that the price of lithiumion batteries could fall dramatically by 2020, creating conditions for the widespread adoption of electrified vehicles in the markets. The aim of this work is to analyze the behavior of titania nanotubes acting as anode for lithium ion batteries. Titania is chemically stable, economically competitive, nontoxic, and an environmental sustainable. Potentiostatic processes were used to treat electrochemically small sheets of commercially pure grade-3 titanium in order to generate titania nanotubes on its surface. After that, the samples were used for to prepare the working electrodes and so tested in a electrochemical cell with lithium metal as anode and LiPF6 in EC-DMC 1:1 as electrolyte. Structural and morphologic characterization of the titania nanotubes have been done by RDX and SEM analysis, while galvanostatic cycles, for to highlight the electrochemical performance as electrodic material, have been made. A comparison of the electrochemical performance with a commercial nanostructured titanium oxide (P25 Degussa), TiO2 obtained by Laser and a commercial lithium titanate Li4Ti5O12 has been made. The TiO2 nanotube electrodes, obtained by our technical anodization, reduces the overall cell voltage and provides good capacity retention on cycling and higher capacity at all used C-rate was delivered.

Generalized min-max classifier

A new neuro-fuzzy classifier, inspired by the min-max neural model, is presented. The classification strategy of Simpsons min-max classifier consists of covering the training data with hyperboxes constrained to have their boundary surfaces parallel to the coordinate axes of the chosen reference system. In order to obtain a more precise covering of each data cluster, in the present work hyperboxes are rotated by a suitable local principal component analysis, so that it is possible to arrange the hyperboxes orientation along any direction of the data space. The new training algorithm is based on the ARC/PARC technique, which overcomes some undesired properties of the original Simpsons algorithm. In particular, the training result does not depend on patterns presentation order and hyperbox expansion is not limited by a fixed maximum size, so that it is possible to have different covering resolutions. A toy problem and two real data benchmarks are considered for illustration.