1,720,990 research outputs found
Riconoscimento automatico di difetti per la diagnostica predittiva su sistemi di isolamento
No full textUna macchina elettrica, durante il ciclo di vita, presenta fenomeni di invecchiamento nel sistema di isolamento. Questi si caratterizzano come impulsi di corrente, conosciuti come "scariche parziali", e sono sia il sintomo sia la causa del deterioramento dell'isolamento stesso. Per questo motivo vengono programmate azioni di manutenzione periodiche per limitare i danni che possono provocare. Risulta però utile monitorare continuamente lo stato di una macchina, sia per non fermarla per una manutenzione non necessaria, sia per valutare online la condizione dell'isolamento, in modo tale da intervenire immediatamente nel caso in cui un difetto grave si manifesti improvvisamente. Le scariche parziali possono essere più o meno pericolose in funzione di alcuni fattori quali l'intensità, la frequenza con cui si manifestano e la posizione all'interno di un motore elettrico. Risulta dunque necessario distinguere la/e sorgente/i che generano tali fenomeni.
Perciò, in questa tesi, vengono presentati diversi approcci e tecniche per il riconoscimento automatico di difetti, sia con algoritmi di apprendimento supervisionato che non. Nel primo caso si identificano soluzioni di apprendimento rapido che possono essere realizzate su dispositivi hardware, con un ottimo compromesso tra capacità di generalizzazione e occupazione di area. Nel secondo, vengono confrontati diversi algoritmi presenti in letteratura e proposta una scelta alternativa dei parametri in ingresso ad essi, che porta a risultati soddisfacenti
Investigating Cutaneous Mechanoreceptors for Neuromorphic Tactile Texture Classification
No full textThis paper investigates the computational cost of modeling the response of the Type-I and Type-II cutaneous human mechanoreceptors for neuromorphic texture classification. We examined both the number of floating operations for modeling the receptors, and the number of synaptic operations for recurrent spiking neural networks (RSNNs) used in classification. Results show that deeper receptors (Type-II) require a greater computational cost to be modeled than those close to the surface (Type-I). However, RSNNs linked with deeper receptors exhibit a lower cost. We evaluated the energy consumption of the modeling and classification parts, each on its dedicated hardware device. The results suggest that pairing Type-I receptors with their corresponding RSNNs offers the best trade-off between energy consumption and classification accuracy
Digital Architecture for the n-mode Tensor-Matrix Multiplication Based on Pipelined Computing Units
No full textCompact digital circuitry supporting data processing is a key requirement of modern engineering. This pa-per addresses the design of digital architectures for a crucial operation in multi-linear algebra: the n-mode tensor-matrix product, implemented in fixed-point representation. A pipelined architecture that optimizes throughput and balances area and energy consumption is proposed. A cost-effective classifier based on this architecture was deployed on an embedded system. Ex-perimental tests conducted on a Kintex-7 FPGA demonstrate that the circuit achieves efficient digital implementations, providing real-time performance on benchmark applications with power consumption lower than 130 mW. This implementation proves to be more efficient than its non-pipelined counterpart
An approximate randomization-based neural network with dedicated digital architecture for energy-constrained devices
No full textVariable energy constraints affect the implementations of neural networks on battery-operated embedded systems. This
paper describes a learning algorithm for randomization-based neural networks with hard-limit activation functions. The
approach adopts a novel cost function that balances accuracy and network complexity during training. From an energyspecific perspective, the new learning strategy allows to adjust, dynamically and in real time, the number of operations
during the network’s forward phase. The proposed learning scheme leads to efficient predictors supported by digital
architectures. The resulting digital architecture can switch to approximate computing at run time, in compliance with the
available energy budget. Experiments on 10 real-world prediction testbeds confirmed the effectiveness of the learning
scheme. Additional tests on limited-resource devices supported the implementation efficiency of the overall design
approac
Hardness Discrimination Using Piezoelectric-Based Biomimetic Tactile Sensor and Machine Learning
No full textIn this letter, we present a tactile sensing system based on piezoelectric sensors, embedded electronics, and a machine learning (ML)-based approach for hardness discrimination. Various statistical features were extracted and evaluated through machine learning algorithms including support vector machines (SVM), single-layer feed-forward neural networks, and k-nearest neighbor (KNN). Five hardness objects were examined by performing indentation experiments using a Cartesian robot equipped with the sensing system while varying the indentation speed and load. Results showed that the SVM classifier trained on features ranked using principal component analysis (PCA) achieves a discrimination accuracy of 96% while utilizing a single sensor. Furthermore, results demonstrated that fixing the indentation speed and load increases the discrimination accuracy to 100%. This study demonstrated the capability of the tactile sensing system in extracting tactile information opening up interesting perspectives for wearable sensing and soft robots
A Digital Implementation of Extreme Learning Machines for Resource-Constrained Devices
No full textThe availability of compact digital circuitry for the support of neural networks is a key requirement for resource-constrained embedded systems. This brief tackles the implementation of single hidden-layer feedforward neural networks, based on hard-limit activation functions, on reconfigurable devices. The resulting design strategy relies on a novel learning procedure that inherits the approach adopted in the Extreme Learning Machine paradigm. The eventual training process balances accuracy and network complexity effectively, thus supporting a digital architecture that prioritizes area utilization over computational performance. Experimental tests confirm that the design approach leads to efficient digital implementations of the predictor on low-performance devices
Comparison Between PD Acquisition System Measurements Using Different Number of Bits for the Quantization
No full textThe PRPDA (Phase Resolved Partial Discharges Acquisition) systems usually employ 8 bits mathrm A/mathrm D converters. Avoiding the converter saturation, it has been necessary, during the measurements, to set the reference voltage on the expected maximum PD signals magnitude. There is an amplitude threshold below which PD signals are not sufficiently quantized to let the separation among PD signals relevant to different defects and noise using automatic recognition algorithms. Therefore, it is necessary to employ PD acquisition systems that exploit more bits for the quantization in order to better distinguish the PD signals from the environmental noise. In this study, PDs and noise measurements have been performed considering for the quantization 8 bit and 12 bit acquisition systems. The efficiency of the recognizing method on the two datasets has been compared and evaluated. The considered algorithm is based upon the Equivalent Time - Equivalent Bandwidth (TW) projection of the data in order to allow a simple clustering in a 2D space
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
- …
