Search

In recent years, deep generative models for graphs have been used to generate new molecules. These models have produced good results, leading to several proposals in the literature. However, these models may have troubles learning some of the complex laws governing the chemical world. In this work, we explore the usage of the histogram of atom valences to drive the generation of molecules in such models. We present Conditional Constrained Graph Variational Autoencoder (CCGVAE), a model that implements this key-idea in a state-of-the-art model, and shows improved results on several evaluation metrics on two commonly adopted datasets for molecule generation.

A novel, low temperature process for the growth of silicon nanowires containing a monocrystalline Si-BC8 phase is presented. Silicon atoms in Si-BC8 phase are arranged in body-centered-cubic unit cells resulting into a different energy-wavefactor diagram compared to the silicon diamond cubic phase. Indeed, Si-BC8 possesses a direct band gap as low as 30 meV at room temperature. These characteristics are highly desirable for a large variety of applications, requiring CMOS-compatible manufacturing. The growth was performed in a CVD reactor under exposure to microwaves, and employing Sn nanospheres and SiH 4 as catalyst and precursor gas, respectively. Microwaves allowed for selective heating of the metal catalyst while keeping the substrate at low temperature. Systematic structural analysis and a phenomenological model for Si-BC8 phase formation are discussed.

It has been widely demonstrated that active network management (ANM) strategies will be required to avoid the violation of network operational limits in distribution networks with a rich presence of distributed energy resources (DERs). With the characteristics of different ANM strategies been varied, e.g. centralized or decentralized, a platform where the benefits and drawbacks of multiple approaches can be easily quantified and benchmarked is required. This work introduces an open-source python-based package with a three-phase unbalanced optimal power flow (OPF) model that is automatically created by extracting the data available from any network modelled in OpenDSS. This OPF formulation is flexible, enabling to incorporate any DER model and modify the objective function and constraints, opening many simulation opportunities.

Motion coordination between the operating table and imaging or surgical devices is becoming increasingly relevant for minimally invasive procedures in the hybrid operating room. In this paper, we presented a mathematical framework for the generation of a virtual remote center of motion (RCM) for a 5 degrees-of-freedom (DOFs) motorized surgical table. The capability of the table to synchronize its orientation to a rotating paired medical device was also addressed. The table inverse kinematic problem was solved by efficient algorithms, which were tested in a virtual simulation environment using a parametrized table model. The proposed solutions support arbitrary RCM position and can be adapted to different 5-DOFs table models by parameter tuning.

As the adoption of residential photovoltaic (PV) continues to increase, its influence on distribution feeder voltage and currents also increases. Effective allocation or modelling on the appearance of PV across the system as a function of an adoption forecast is an important consideration for future distribution planning. The spatial information required for the forecast/allocation process is expected to be available to utilities at a cost proportional to its level of detail. Naturally, there is a need to understand the potential trade-offs between different modelling approaches and options. This paper explores three allocation models that differ on the complexity of their allocation mechanism. Both the error due to ignoring customers' PV adoption mechanisms and due to the PV forecast uncertainty are explored and compared.

This paper has been presented at: 2019 IEEE Milan PowerTech In this paper a methodology is proposed to calculate the load demand flexibility that could be activated within the next 24-hours for solving the technical impacts of contingencies that may come up in an unbalanced low voltage distribution networks with high penetration of intermittent DG sources. The methodology is formulated within a Demand Response program environment via load shifting as flexibility enabler mechanism. To achieve that, a non-linear optimisation problem is formulated based on an unbalanced optimal power flow, which allows the determination of the load flexibility that each Demand Response customer could provide at the request of the Distribution System Operator. The demand as well as weather conditions are forecasted for the day ahead. The optimisation problem is solved in a sequence fashion, within a daily framework, splitting the whole problem in optimisation blocks. In each block, the flexible load demand is obtained and the load demand forecasting its updated for the upcoming blocks based on the changes in the scheduled load demand. The methodology is applied to a real distribution network with the load data received from the smart metering infrastructure. The results obtained show the strength of the methodology in solving the technical problems of the network under high unbalanced operation. The work done by Valentin Rigoni and Alireza Soroudi is supported by a research grant from Science Foundation Ireland (SFI) under the SFI Strategic Partnership Programme Grant No. SF/15/SPP/E3125. The opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Science Foundation Ireland. Jose Angel Velasco and Hortensia Amaris are supported by the Spanish Ministry of Economy and Competitiveness through the National Program for Research under the project OSIRIS (RTC-2014-1556-3).

The US Army Aviation and Missile Research, Development, and Engineering Center (AMRDEC), in collaboration with the US Army Engineer Research and Development Center (ERDC), is using machine learning (ML) to transform the way rotorcraft maintenance logbook event data is scored for reporting purposes. Traditionally, human analysts manually inspect data fields completed by maintenance personnel and, using published guidelines in conjunction with their personal expertise, provide sets of labels or scores for each event. These labels are stored with the data and provide valuable insight into maintenance event histories. However, the inequity between the enormous volume of maintenance data generated daily and the ability of analysts to score the data results in only 10% of all data receiving scores; therefore, 90% of the recorded data does not contain this important value added feature. Classification algorithms for automating this scoring process trained on existing labeled data sets have been implemented with promising results. A particularly challenging element of this problem, however, involves the classification of the specific component on which maintenance was performed. Greater than 1200 unique labels exist that can be used to describe a rotorcraft component that is the subject of a maintenance action. Furthermore, the component labels are hierarchically structured, resulting in the occurrence of multiple levels of precision in identifying a component in the expert-labeled data used for training. Although computational efficiency of common classification algorithms has improved considerably, it is still quite challenging to harness these methods for problems that include large numbers of unique class labels. This paper describes several novel strategies for solving this problem, based on hierarchical ensemble models and strategic label set segmentation. Through these approaches, a best overall total component classification accuracy of 96% was achieved, in conjunction with a total per record accuracy for three different label categories of 93%. The approaches implemented to handle the large label set for classification of rotorcraft components, along with classification performance measures, are discussed.

With increasing penetration levels of photovoltaic panels (PVs) in low voltage (LV) networks, voltage statutory limits are likely to be violated. To avoid this, most strategies rely on the capacity of inverters to modify power injections as needed. In cases for which the required set points are not obtained from a constrained formulation of the power flow problem, voltage sensitivities to power injections are required. Therefore, recent publications have proposed methods to estimate these sensitivities in residential LV feeders. However, as they rely on extensive monitoring and communication infrastructure, they may result impractical for operators to implement. This paper introduces a novel methodology that allows every PV to autonomously estimate voltage sensitivities based merely on the real-time local voltage at the customer point of connection. A robust accuracy assessment under multiple system states is performed and benchmarked against a more traditional method.

Received Signal Strength Indication (RSSI) is commonly used to provide distance estimates in range-based localization. In most cases, the localization systems use RSS at short range where the distance estimates are more reliable or use RSS alongside other techniques such as Time of Flight (ToF). This is so, since RSSI measurements have relatively high variance at long range and are strongly influenced by occlusions and interference in the deployment region of the Radio Frequency (RF) devices. This paper presents an overview of common filtering techniques that can be used to process RSSI readings in order to improve the accuracy of range computation from raw RSSI with minimal computational overhead. The range estimates computed from the filtered data are compared with expected values of the perturbed range/distance expressed in terms of the Cramer-Rao Lower Bound (CRLB) for RSS distance estimation. Results show that filtering can significantly improve the accuracy of range estimation, highlighting the pros and cons of the presented filtering methods at different range values.

The present and future data management requirements for fusion experiments are presented along with the currently adopted solutions. Even if the presented solution fulfil the requirements of the current experiments, the next generation fusion devices are likely to produce/require an unpreceded amount of data. For this reason, the solutions adopted nowadays, and also foreseen for the experiments under construction, might prove not enough scalable. Information Technology already provides efficient solutions for big data management, successfully employed for large cloud applications and social media. In particular, MongoDB, Cassandra and Hadoop represent promising candidates for the next generation experiments because their combined usage covers the specific data requirements for fusion research.

There is need for reliable, low cost detectors to continuously monitor the air quality in presence of hazardous gases, harmful for human health. In this work we demonstrate how the use of different type of metal oxide one-dimensional sensors based on ZnO and SnO 2 could provide high sensing capabilities together with differentiated responses to gases. The sensors can then be used together into a cost-effective portable sensor systems for quantitative detection of harmful emissions and contaminations. The gases tested in the present work are H 2 a highly explosive gas (4% is the lower explosion limit), H 2 S, one of the principal compounds involved in the natural cycle of sulphur in the environment, and DMMP, a simulant of the nerve agent Sarin. ZnO and SnO 2 nanowires were prepared by Physical Vapour Deposition and ZnO polycrystalline nanorods were prepared by RF sputtering. The optimum response to low concentrations of DMMP at 500°C was observed for ZnO materials, with respect to SnO 2 at 500°C. At 400°C H 2 is better sensed by ZnO. Detection of H 2 S is better obtained with ZnO NRs compared to the other sensors.

On-load tap changer-fitted transformers have been proposed to solve voltage problems in low voltage (LV) networks with rich penetration of distributed generation. However, knowledge of the voltage level at customers' point of connection is key for the performance of potential control strategies. This work proposes a generic methodology to estimate voltages in LV feeders without the need of remote monitoring. The methodology relies on local flow measurements at the transformer level and residential load models. Maximum Likelihood Estimation (MLE) is applied to obtain the most likely residential loads' power consumptions which are later used to calculate the associated nodal voltages. A novel formulation of power flow equations based on sensitivity analysis is used to simplify the MLE problem. The methodology is tested on a real unbalanced feeder with unique presence of residential loads. The high accuracy of results promotes it as a potential alternative to monitoring investments.

In this paper, we present a framework for detecting tampered information in digital videos. Using the proposed technique is possible to detect several types of tampering with a pixel granularity. The framework uses a combination of temporal and spatial watermarks that do not decrease the perceived quality of the host videos. We use a modified version of Quantization Index Modulation (QIM) algorithm to store the watermarks. Since QIM is a fragile watermarking scheme, it is possible to detect local, global, and temporal tampers and also estimate the attack type. The framework is fast, robust, and accurate.