Search

In this work, we introduce a new architecture to address the Visual Question Answering problem, an open field of research in the NLP and Vision community. In the last few years, with the advent of Deep Learning and the exponential growth of computing power, researches came up with brilliant solutions to tackle the problem. However, most of the related work share a standard limitation: the number of possible answers is usually restricted to a limited set of candidates, limiting the power of such models. In this work, we describe a new architecture that employs new state-of-the-art language models, such as the Transformer, to generate open-ended answers. In the end, our contribution to the scientific community lies in a new approach that allows VQA systems to generate unconstrained answers. First, we introduce the necessary background as well as the most critical computational models to deal with text and images. Ultimately, we show that our architecture compares well with other VQA models, setting a new baseline for future work.

An audio power amplifier achieved with a hybrid architecture constituted by the magnetic coupling of a linear amplifier (Hdlin) with a switching amplifier (Hdsw) placed on two coaxial coils of a loudspeaker (LS1), or with a hybrid architecture constituted by the electrical coupling of a linear amplifier (Hdlin) with a switching amplifier (Hdsw). These solutions allow obtaining an amplifier characterized by the low distortion of a linear amplifier and by the high efficiency of the switching amplifier.

Un sistema di elaborazione numerica in grado di effettuare una scomposizione di un segnale in bande di analisi per ottave e terzi di ottave e calcolare l’energia relativa al segnale in ogni sigola banda.Il sistema si basa su una serie di filtri digitali di ultima generazione, che opportunamente configurati ed inseriti in una struttura ricorsiva, consentono di ottenere analisi di frequenza in ampie bande con una contenuta complessità computazionale ed una ridotta richiesta di memoria.Inoltre la specifica ottimizzazione consente di implementare tale algoritmo anche su processori con architettura in ‘virgola fissa’.

Open-loop power converters are feasible if the output resistance of the converter is as low as possible, and a possible solution is the minimization of power losses. The solution is the use of a three-phase transformer with a delta-wye connection within a full-bridge converter topol- ogy. The configuration will be referred to as 3boost power supply. The three-phase transformer replaces the common single-phase transformer, and it is driven by a three-phase full-bridge inverter operating in six-step modulation. At secondary, a three-phase full wave diode rectifier is used to obtain the output dc volt- age level. Therefore, a unitary transformer utilization factor is achieved. This topology allows to achieve a high efficiency, a lower ripple factor, and a good EMC behavior.

In this paper, a diagnostic procedure for rotor bar faults in induction motors is presented, based on the Hilbert and discrete wavelet transforms. The method is compared with other procedures with the same data, which are based on time–frequency analysis, frequency analysis and time domain. The results show that this method improves the rotor fault detection in transient conditions. Variable speed drive applications are common in industry. However, traditional condition monitoring methods fail in time-varying conditions or with load oscillations. This method is based on the combined use of the Hilbert and discrete wavelet transforms, which compute the energy in a bandwidth corresponding to the maximum fault signature. Theoretical analysis, numerical simulation and experiments are presented, which confirm the enhanced performance of the proposed method with respect to prior solutions, especially in time-varying conditions. The comparison is based on quantitative analysis that helps in choosing the optimal trade-off between performance and (computational) cost.

Condition monitoring of electric machines is a topic of increased interests for the quality constraint of many applications. Mature technologies exist for stationary conditions, that can achieve nice results at incipient stage without affecting machine normal operation. However, there are still open research issues for time-varying conditions, and variable loads. Hence, a testbed for fast development of diagnostic methods is beneficial to research activities in this field. Here, a testbed was designed and realized, that allow to test different methods in real time, including suitable signal processing. The testbed was tested for rotor fault detection in induction machines. Results show that the testbed can be used for real-time validation of diagnostic methods.

Condition monitoring of electric machines is gaining interest in industry, because of increasing demand of fault tolerance machines. State-of-the-art diagnostic procedure are based on non-invasive signal processing of electrical signal that allow to detect fault signature at an incipient stage. Here, the use of Hilbert and Wavelet transform is investigated. Specifically, a theoretical analysis is presented that can be used to select the optimal wavelet, i.e. the decomposition level that corresponds to the maximum fault signature energy. Simulation results confirm the effectiveness of the proposed procedure, even under time-varying conditions.