Laboratoire des Sciences du Numérique de Nantes

The Interest Public Group ARRONAX's C70XP cyclotron, used for radioisotope production for medical and research applications, relies on complex and costly systems that are prone to failures, leading to operational disruptions. In this context, this study aims to develop a machine learning-based method for early anomaly detection, from sensor measurements over a temporal window, to enhance system performance. One of the most widely recognized methods for anomaly detection is Isolation Forest (IF), known for its effectiveness and scalability. However, its reliance on axis-parallel splits limits its ability to detect subtle anomalies, especially those occurring near the mean of normal data. This study proposes a hybrid approach that combines a fully connected Autoencoder (AE) with IF to enhance the detection of subtle anomalies. In particular, the Mean Cubic Error (MCE) of the sensor data reconstructed by the AE is used as input to the IF model. Validated on proton beam intensity time series data, the proposed method demonstrates a clear improvement in detection performance, as confirmed by the experimental results.

Despite recent advances in operational streamflow forecasting systems, anticipating and forecasting droughts and associated low-flow conditions remain major challenges in hydrology, with substantial impacts on water-dependent sectors such as agriculture and industry. Enhancing sub-seasonal to seasonal streamflow forecasting is therefore critical for improving water resources management. This study investigates the performance of a Handoff forecast Long Short-Term Memory (LSTM) architecture (Nearing et al., 2024) for probabilistic streamflow forecasting at lead times extending up to six months, with a particular emphasis on low-flow conditions.The Handoff forecast LSTM is trained regionally on a subset of 292 basins from the Catchment Attributes and MEteorology for Large-sample Studies - FRance dataset (CAMELS-FR) (Delaigue et al., 2025), after excluding basins affected by unreliable low-flow measurements. Model training relies on basin-averaged hydro-meteorological reanalysis data provided by CAMELS-FR. Evaluation of the model is conducted using ensemble streamflow forecasts generated from historical scenarios and meteorological ensemble predictions from the SEAS5 model from the European Center for Medium-Range Weather Forecasts (ECMWF) (Johnson et al., 2019)The generated ensemble streamflow forecasts are evaluated using a set of probabilistic metrics such as the Continuous Ranked Probability Score (CRPS), the Brier Score, the Area under the ROC curve, and the Talagrand diagram, and using the natural streamflow climatology as a reference. In addition, a sensitivity analysis of static catchment attributes is performed to assess their relative contribution to model performance and to better understand the drivers of predictability across basins.Delaigue, O., Guimarães, G. M., Brigode, P., Génot, B., Perrin, C., Soubeyroux, J.-M., Janet, B., Addor, N., & Andréassian, V. (2025). CAMELS-FR dataset: a large-sample hydroclimatic dataset for France to explore hydrological diversity and support model benchmarking. Earth System Science Data, 17(4), 1461–1479. https://doi.org/10.5194/essd-17-1461-2025Johnson, S. J., Stockdale, T. N., Ferranti, L., Balmaseda, M. A., Molteni, F., Magnusson, L., Tietsche, S., Decremer, D., Weisheimer, A., Balsamo, G., Keeley, S. P. E., Mogensen, K., Zuo, H., & Monge-Sanz, B. M. (2019). SEAS5: The new ECMWF seasonal forecast system. Geoscientific Model Development, 12(3), 1087–1117. https://doi.org/10.5194/gmd-12-1087-2019Nearing, G., Cohen, D., Dube, V., Gauch, M., Gilon, O., Harrigan, S., Hassidim, A., Klotz, D., Kratzert, F., Metzger, A., Nevo, S., Pappenberger, F., Prudhomme, C., Shalev, G., Shenzis, S., Tekalign, T. Y., Weitzner, D., & Matias, Y. (2024). Global prediction of extreme floods in ungauged watersheds. Nature, 627(8004), 559–563. https://doi.org/10.1038/s41586-024-07145-

Benzene, toluene, ethylbenzene, and xylene (BTEX) are hazardous air pollutants with high toxicity and a strong potential for secondary pollutant formation. However, their occurrence and behavior in the Arctic remain poorly understood. During the Alaskan Layered Pollution and Chemical Analysis (ALPACA) field campaign in Fairbanks, Alaska in January-February 2022. Surface observations in downtown Fairbanks revealed two major pollution periods, with extremely cold (down to -35°C) and warmer temperatures (around 0°C), respectively. BTEX concentrations reached 4–12 times higher than those reported in the US and European countries under dark, cold Arctic winter conditions at breathing level, posing a significant health risk. We simulated BTEX atmospheric distributions in the Fairbanks region using the FLEXible PARTicle-Weather Research and Forecasting (FLEXPART-WRF) Lagrangian particle dispersion model and anthropogenic emissions at the surface and aloft. Due to limited photochemical loss in to the dark polar winter conditions, we treat BTEX as an unreactive tracer in the model. The control run with the emission inventory developed by Alaska Department of Environmental Conservation (ADEC) substantially underestimates BTEX concentrations compared to observations during both polluted periods, indicating deficiencies in winter emissions and near-surface mixing. Enhancing cold-start gasoline vehicle emissions by a factor of 2 during very low-temperatures substantially improved model results during the cold polluted period, while introducing a relative humidity dependence for mobile emissions improved simulated BTEX during the warm, humid pollution period. Addition of emissions of residential heating oil aromatics, not taken into account in the ADEC inventory, also reduced normalized mean biases by 5-10%.The improved model simulation was used to investigate contributing source sectors. While mobile traffic emissions were identified as the dominant source of BTEX across the Fairbanks North Star Borough, residential heating and non-point sources contributed substantially in downtown Fairbanks. Replacing residential wood burning in the inventory with oil heating during severe pollution periods, in line with air quality control guidelines, was found to effectively reduce BTEX concentrations, particularly benzene by up to 30%. While persistent surface-based temperature inversions largely confined BTEX below ~20 m, upward transport, induced by wind shear, during severe episodes, sometimes lofted near-surface pollutants to higher altitudes, potentially contributing to regional pollution and background Arctic haze.The findings of this study emphasise the need to accurately account for temperature and humidity dependent vehicle emissions, residential oil heating emissions, and winter boundary-layer dynamics for improved simulations of air quality in cold wintertime environments, not only in the Arctic but also in mid-latitudes.

Chenouard et al., (2026). RealPaver 1.1: A C++ Library for Constraint Programming over Numeric or Mixed Discrete-Continuous Domains. Journal of Open Source Software, 11(118), 9331, https://doi.org/10.21105/joss.09331

This paper presents the development and experimental evaluation of the Athena parallel robot, a novel system designed for robot-assisted pancreatic surgery. The development of the experimental model based on the kinematic scheme, including the command and control system (hardware and software), the calibration procedure, and the performance measurements of the experimental model based on finite element analyses of the 3D model, are also detailed in this paper. Based on these finite element analyses, a region of the robot that introduces clearance during the operation of the experimental model is found. The paper also presents the methodology used for mapping the robot’s workspace with an optical system, which enabled improvements to ensure coverage of the entire pancreas area. The results obtained before and after the mechanical improvements are presented, demonstrating a reduction in clearance by up to 4.1 times following part replacement, as well as a workspace extension that enables the active instrument to reach the entire pancreatic region.

This letter introduces a novel approach for graph classification in the context of sensor network data analysis using machine learning methods. It is based on exploiting both sensor positions and their functional connectivity for the inference of a graph class. The original idea is to formulate interaction forces between the graph vertices and exploit them as features of the graph functional connectivity. This letter also discusses the fusion of these features with classical ones extracted from sensor signals and structural features of the learned underlying graph. This original framework is found to be relevant for machine learning tasks, for instance the classification of multivariate EEG signals, where the force feature representation allows to distinguish between different mental workload levels. Experiments conducted with real data illustrate the usefulness of the proposed framework.

Initially designed for Knowledge Graphs (KG) containing plain (s, p, o) triples, research on link prediction has evolved to address more complex structures such as hypergraphs, hyper-relational graphs, and bi-level graphs. These advancements push link prediction to handle annotation patterns where a triple is annotated with another (t, p, t), or where a triple is annotated with qualifiers (t, p, o). However, each approach focuses on a single annotation pattern, and the case where a triple is annotated with a subject (s, p, t) has never been explored. In this paper, we propose LP4P, the first link prediction model capable of predicting entities in both plain triples and three annotation patterns. LP4P captures the information expressed in annotations through a four-pattern attention layer, and its loss function further leverages the ontological information of KGs. Moreover, we built WD4P, an RDF KG derived from baseline datasets, which includes the four patterns. Extensive empirical evaluations demonstrate that LP4P outperforms relevant state-of-the-art models on knowledge graphs with the four patterns, and achieves comparable performance on standard benchmarks.