Higher Institute on Territorial Systems for Innovation
PORTO@iris (Publications Open Repository TOrino - Politecnico di Torino)Not a member yet
146173 research outputs found
Sort by
On Traffic Matrix Estimation via Super-Resolution and Federated Learning
No full textNetwork traffic telemetry plays a crucial role in the
management of modern networks. Estimation of the network
traffic matrix is a widely recognized problem whose solutions
can span a diverse set of applications. Current approaches
to traffic matrix inference through statistical methods often
rely on assumptions about the matrix structure, which may
be invalid in certain scenarios. Data-driven methods, instead,
often use detailed information about the network topology that
may be unavailable or impractical to collect. To overcome these
challenges, we propose a super-resolution technique for traffic
matrix inference that leverages coarser measurements to predict
fine-grained network traffic. Furthermore, we devise a distributed
learning procedure and adapt our model to scenarios of partial
network visibility. Our experiments on real network traces
demonstrate that the proposed approach can infer fine-grained
network traffic with high precision. Moreover, we prove that
our distributed approach improves the inference accuracy with
respect to its centralized counterpart, significantly lowering the
training time, even in scenarios with partial network knowledge
Soil Structure and Mixing Controls on Water‐Rock Contact: Implications for Enhanced Weathering
Get PDFEnhanced weathering (EW), the addition of finely ground silicate rock powder (RP) to soil, has
emerged as a promising carbon removal strategy. However, quantifying weathering rates in soils remains
challenging, as most continuum‐scale EW models do not adequately account for the fraction of RP surface area
(SA) that is wet at a given soil moisture and thus actively weathering. Here, we study how soil pore structure, RP
particle size distribution, and RP mixing degree within the soil control water‐rock contact. Using a soil‐physics‐
based framework, we derive a scaling factor that quantifies the wet fraction of RP SA as a function of soil
moisture and mixing degree within soil pores. This scaling factor varies nonlinearly with soil moisture for
typical soil pore structures and RP particle size distributions, countering previous zero‐order (independent of
soil moisture) or linear assumptions. The scaling factor evolves dynamically with hydrological fluctuations and,
for a given pore structure and RP mixing degree, it can span nearly two orders of magnitude with changes in
median particle size. To illustrate its application, we integrate the derived scaling factor into the Soil Model for
Enhanced Weathering and examine the sensitivity of simulated weathering fluxes to mixing degree under
otherwise identical conditions. Under low mixing, results show that average weathering rates are roughly two
orders of magnitude lower than under perfect mixing over 1 year of application. Our work provides a
mechanistic, computationally efficient framework for representing water‐rock contact in soil, offering a
pathway to improve continuum‐scale EW models
A label-free hybrid fault detection and diagnosis approach for HVAC systems using bayesian networks
Get PDFThe building sector accounts for up to 40 % of global energy demand, with HVAC systems responsible for nearly half of this consumption. Faulty HVAC operation can result in substantial energy waste, reduced equipment lifespan, and increased operational costs. To address these challenges, the scientific community has focused on developing Fault Detection and Diagnosis (FDD) strategies that are both accurate and applicable in real conditions. Although data-driven approaches have shown strong potential, their practical deployment remains limited by the need for labeled data and variables not commonly available in real buildings.
This paper presents a hybrid FDD framework based on Bayesian Networks (BNs) that combines data-driven models with expert knowledge. A set of reference models, using Random Forest algorithms, was developed to predict key variable values and define a baseline. Deviations from this baseline, expressed as residuals, were converted into virtual evidence and combined with hard evidence derived from domain expertise. These inputs were fed into a set of BN models, one for each operational mode, whose parameters and system-level structure were informed by expert knowledge and constructed efficiently through semantic metadata schemas based on brick ontology.
The BNs performed fault detection and component isolation while the diagnosis is supported by targeted statistical analyses tailored to each system component. The main advantage of the proposed framework is that it requires only variables typically available in building management systems and does not rely on the a-priori knowledge of fault labels.
The approach was validated on a simulated dataset from the Single Duct Air Handling Unit developed by the LBNL and further tested on a Fan Coil Unit. It achieved detection and isolation accuracies of approximately 91 % and 87 %, respectively, confirming its robustness, adaptability, and practical relevance
Orchestrating Composite Applications at the Edge
Get PDFEdge computing plays a pivotal role in enabling
time-critical Machine Learning (ML) by bringing computational
capabilities closer to end users. However, satisfying stringent
inference latency and quality constraints under varying task
complexity and limited edge resources remains challenging. We
tackle this by proposing a novel architectural approach for ML-
based edge deployments and introducing CARE, our orchestra-
tion framework that jointly configures composite applications and
compute resources to meet inference latency and quality targets,
while minimizing energy consumption. Experimental results in
the context of Multi-Object Tracking (MOT) demonstrate that
CARE improves inference quality by up to 50% and reduces
latency by up to 2× over monolithic baselines
A Multiterminal Setup for Complex Dynamics Characterization and Unconventional Computing in Self-Organizing Memristive Networks
Get PDFThe recent growing interest in neuromorphic architectures based on emergent dynamics of self-organizing memristive networks has posed some challenges regarding the spatiotemporal characterization of these multiterminal systems. This work presents a versatile measurement platform specifically designed for the characterization of memristive nanowire networks and for testing the implementation of unconventional computing paradigms in these systems. By integrating an FPGA controlled, parallel multiterminal array of source-measure units with a custom fixture based on spring-loaded electrodes, the system allows for real-time, reconfigurable voltage and current measurements across 16 terminals without hardware reconnections. The platform supports seamless transition between conventional two-terminal characterization, multiterminal characterization and testing computational properties in the framework of physical reservoir computing. Local conductance measurements, voltage mapping, and real-time dynamic monitoring offer unique insights into the spatiotemporal behavior of the networks. Furthermore, we show that the system enables to correlate electrical properties of the multiterminal network in terms of conductance matrices and voltage maps with computational performances, allowing also adaptive control over the network's operating state. The here reported setup provides a versatile platform for computing at the matter level (i.e., in materia) with multiterminal systems based on self-organizing memristive networks
GLEm-Net: Unified framework for data reduction with categorical and numerical features
Get PDFIn an era of effortless data collection, the impact of machine learning — especially neural networks (NNs) — is undeniable. As datasets grow in size and complexity, efficiently handling mixed data types, including categorical and numerical features, becomes critical. Feature encoding and selection play a key role in improving NN performance, efficiency, interpretability, and generalisation. This paper presents GLEm-Net (Grouped Lasso with Embeddings Network), a novel NN-based approach that seamlessly integrates feature encoding and selection directly into the training process. GLEm-Net uses embedding layers to process categorical features with high cardinality, simplifying the model and improving generalisation. By extending the grouped Lasso regularisation to explicitly consider categorical features, GLEm-Net automatically identifies the most relevant features during training and returns them to the analyst.
We evaluate GLEm-Net on open and proprietary industry datasets and compare it to state-of-the-art feature selection methodologies. Results show that GLEm-Net adapts to each dataset by allowing the NN to directly select subsets of most important features, offering on par performance with the best state-of-the-art feature selection methods, while eliminating the need for the external feature encoding and selection steps that are now incorporated in the NN training stage
Speech Recognition with Cochlea-Inspired In-Sensor Computing
Get PDFTraditional speech recognition methods rely on software-based feature extraction that introduces latency and high energy costs, making them unsuitable for low-power devices. A proof-of-concept demonstration is provided of a bioinspired tonotopic sensor for speech recognition that mimics the human cochlea, using a spiral-shaped elastic metamaterial. The measured modal response of the structure at different frequencies generates a spatially distributed signal, providing a spatiotemporal map of the input named "tonogram". The device acts as an in-sensor physical reservoir computing system, working simultaneously as a sensor and as a computing unit, capable of extracting features of spoken words relevant to speech recognition. Results indicate that this can serve as a valid alternative to traditional software-based digital preprocessing, ensuring high accuracy in terms of classification, while reducing computational requirements. This work demonstrates the potential of bioinspired metamaterials for energy-efficient auditory sensing and, beyond speech recognition, for applications such as IoT devices and edge computing artificial intelligence systems
Cross-Linking PEG Microgels with Mesoporous Organosilica Nanoparticles to Engineer Microporous Annealed Particle Scaffold Properties
Get PDFMicroporous annealed particle (MAP) scaffolds composed of cross-linked microgels are emerging as important biomaterials due to their modular design, high porosity, and large surface area, which allow for highly tailored properties, cellular infiltration, nutrient and waste exchange, and improved drug release characteristics. To further expand on these technologies and enhance their modular design, thiolated mesoporous organosilica nanoparticles (ssMSN-SH), which degrade in response to glutathione, were first cross-linked inside of 4-arm poly(ethylene glycol) maleimide microgels. The microgel size, rheological properties, and payload release characteristics were determined. Hierarchical MAP scaffolds were then created by cross-linking 4-arm poly(ethylene glycol) maleimide microgels with ssMSN-SH. Surface attachment and cross-linking of microgels using ssMSN-SH were confirmed with scanning electron microscopy and confocal imaging; rheological properties were characterized; porosity was analyzed; glutathione-induced degradation of MAP scaffolds was determined; and the depth of viable NIH/3T3 fibroblasts into MAP scaffolds was visualized using confocal microscopy. ssMSN-SH cross-linked microgels form MAP scaffolds with an increase in the storage modulus, maintain their porosity while decreasing the interparticle distance, degrade in response to glutathione, and increase the depth of viable NIH/3T3 fibroblasts into scaffolds when compared with bulk PEG hydrogels
Optimal dispatch of wind-powered solid oxide electrolyzer plants
No full textSolid oxide electrolyzer (SOEL) systems represent a promising hydrogen production technology for renewable energy plants, offering significant economic benefits due to their best-in-class efficiency. To fully harness their potential in dynamic scenarios, reliable dispatch strategies are essential. In this challenging context, this study develops a dispatch optimization framework for wind-electrolysis plants aimed at maximizing profitability. A key aspect of the framework is a detailed scheduling model of the SOEL system, which captures its nonlinear energy conversion behavior, three-state operation and ramp rate constraints. Notably, implementing a 1-week time horizon and a 5-minute resolution allows the model to account for both cold and warm startup durations of the SOEL. Simulations are conducted using 10 typical weeks of onshore wind generation and day-ahead electricity price data from Germany in 2019. Results indicate that the SOEL system exhibits a relatively stable operational pattern, predominantly following wind generation when in operation while occasionally switching to standby in response to extreme conditions, such as peak electricity prices or low wind power levels. Additionally, a comparative analysis between SOEL and proton exchange membrane electrolyzer (PEMEL) systems is conducted across technical, energy and economic metrics. The impact of varying hydrogen prices on dispatch outcomes is also investigated. Findings suggest that the PEMEL system exhibits greater dynamic behavior than the SOEL, with up to 13 times the total number of state transitions. However, despite its higher investment cost and shorter lifespan, the SOEL case shows a higher profitability, with an average net profit difference of 0.53 k€ and 2.01 k€ across all typical weeks under the hydrogen prices of 3 €/kg and 5 €/kg, respectively