HAL portal of the University of Lorraine
Not a member yet
161261 research outputs found
Sort by
Phi-FEM-FNO: a new approach to train a Neural Operator as a fast PDE solver for variable geometries
No full textInternational audienceIn this paper, we propose a way to solve partial differential equations (PDEs) by combining machine learning techniques and the finite element method called phi-FEM. For that, we use the Fourier Neural Operator (FNO), a learning mapping operator. The purpose of this paper is to provide numerical evidence to show the effectiveness of this technique. We will focus here on the resolution of two equations: the Poisson-Dirichlet equation and the non-linear elasticity equations. The key idea of our method is to address the challenging scenario of varying domains, where each problem is solved on a different geometry. The considered domains are defined by level-set functions due to the use of the phi-FEM approach. We will first recall the idea of -FEM and of the Fourier Neural Operator. Then, we will explain how to combine these two methods. We will finally illustrate the efficiency of this combination with some numerical results on three test cases. In addition, in the last test case, we propose a new numerical scheme for hyperelastic materials following the phi-FEM paradigm
Comparison of silver zeolites of natural and synthetic origins on their ability to trap irreversibly CH<sub>3</sub>I in conditions representative of severe nuclear accident
No full textInternational audienceIn the present work, a low-cost silver natural zeolite (AgN) with clinoptilolite structure and ≈ 9 wt% Ag was specifically designed and evaluated for the capture of methyl iodide (CH3I) in a range of conditions representative of severe nuclear accident (T = 80-140°C, high steam contents, presence of CO, γ-irradiation….). The physicochemical and textural characteristics of AgN were first characterized by SEM-EDX, DR-UV-Visible spectroscopy, N2 porosimetry, elemental analysis and compared to those of well-known synthetic silver faujasite zeolites (AgX and AgY). Common features and differences existing between AgY and AgN in terms of CH3I trapping mechanism were addressed by FTIR spectroscopy and DRIFTS. CH3I adsorption capacities were also computed from CH3I breakthrough curves measured in absence/presence of pre-adsorbed water. Interestingly, the CH3I trapping efficiencies measured at 140°C were found to strongly depend on the type of silver zeolite and steam content. Up to a water content of 42 wt% in the feed, both AgN sorbent and synthetic exchanged faujasites AgX and AgY performed very well, with decontamination factors in the range 5.105-103. At higher humidity (45 wt%), the filling of pores by liquid water induced a rather drastic drop in measured DF values, which is less pronounced for AgY. Additional challenging tests (presence of CO, trapping stability under γ-irradiation) have confirmed the potential of tailored silver zeolites for effective CH3I retention in severe accident scenarios
‘That’s Jazz Made in Germany’ : Émancipation ou contrôle ? Deux scènes berlinoises face à la Guerre froide
No full textInternational audienc
Radiotherapy in the management of PDAC, from past to present
No full textInternational audiencePancreatic ductal adenocarcinoma (PDAC) remains one of the most lethal forms of cancer, with a dismal survival rate. Therapeutic options are restricted in surgery for patients with resectable disease and in chemotherapy for those with unresectable disease. The potential benefits of radiotherapy (RT) or chemoradiotherapy (CRT) have been extensively investigated in both neoadjuvant and adjuvant settings in the management of patients with PDAC. Nevertheless, a substantial number of clinical trials have yielded conflicting findings, thereby rendering the impact of RT on patient survival and margin-negative (R0) resection inconclusive. A comprehensive examination of the historical evolution of RT in PDAC, encompassing the identification of both constraints and opportunities for advancement, is essential to establish RT as a promising therapeutic avenue for patients with PDAC. The aim of this review is to provide a synthesis of past clinical trials and future studies to elucidate the evolving role of RT in the management of PDAC as adjuvant and neoadjuvant CRT across different stages of the disease
Z-scheme BiOBr-BiOI on bentonite for boosted visible-light degradation of cefazolin and rhodamine B
No full textInternational audiencePhotocatalysts performance is often restricted by low specific surface area, rapid electron–hole recombination, and a wide bandgap. To overcome these drawbacks, a novel heterojunction photocatalyst was developed by integrating exfoliated bentonite (BTex) with a BiOBr-BiOI binary composite (BTex@BiOBr-BiOI), prepared at an optimized BiOBr/BiOI molar ratio. Exfoliation of bentonite was achieved using cetyltrimethylammonium bromide (CTAB) in a hot aqueous medium under ultrasonic treatment, providing a high-surface-area support. The composites were characterized by XRD, SEM, EDS, BET, UV–Vis DRS, XPS, and photoluminescence (PL), confirming successful heterostructure formation. Photocatalytic activity was evaluated under visible-light for cefazolin (CFZ) and rhodamine B (RhB) degradation. The BTex@BiOBr-BiOI composite also demonstrated versatility against diclofenac (DFC), amoxicillin (AMX), and Cr(VI) photoreduction. Degradation efficiencies reached 95 % for DFC, 58 % for CFZ, 38 % for AMX (10 mg L-1 each), 100 % for RhB (44 mg L- 1), and 62 % for Cr(VI) (10 mg L-1) within 60 min. The enhanced performance was attributed to synergistic Z-scheme charge transfer between BiOBr and BiOI, along with improved charge- separation promoted by BTex. Reactive species trapping experiments confirmed superoxide radicals (•O2- ) as the dominant species, while holes (h+) and hydroxyl radicals (•OH) contributing secondary roles
Prosumers matching and energy routing through Yens and SOS algorithms in P2P energy trading systems
No full textInternational audienceGrowing clean-energy adoption integrates renewables into grids, enabling peer-to-peer energy trading, where the Energy Internet uses routers and efficient routing algorithms to deliver power between trading pairs. Energy routing involves pairing prosumers and choosing a non-congested, efficient transmission path. Given the unavoidable exogenous grid costs associated with peer-to-peer trading, an energy routing algorithm that accommodates these costs is essential. This paper formalizes the energy routing as a nonconvex mixed-integer nonlinear optimization problem that minimizes the consumer energy cost. To solve it, a semi-decentralized energy routing approach incorporating graph theory and metaheuristics is introduced. It efficiently determines for consumers the cost-effective producers, power allocation, and the efficient energy transmission paths with the lowest energy transmission cost while respecting the grid’s physical and market constraints. These constraints encompass exogenous costs, capacity limits, power flow direction constraints, and power losses accurate calculations during transmission—factors often overlooked by existing energy routing algorithms. It involves an energy transmission scheduling mechanism that addresses path and source conflicts, preventing congestion during simultaneous transmissions. The computational load is balanced between energy routers and network system operator. Simulations demonstrate its effectiveness in solving energy routing, preventing physical constraint violations, resolving path and source conflicts, and optimizing energy costs
Tuning photochemical and biological properties of nitro-nitrosyl ruthenium complexes via nitrogen heterocycle geometry: Benzimidazole vs. Indazole
No full textInternational audienceTwo new nitrosyl ruthenium complexes [RuNO(L) 2 (NO 2 ) 2 OH] (where L = indazole (HInd), benzimidazole (Benz-im)) were synthesized and characterized. The XRD structural data were obtained for acetate salt of the indazole complex -[RuNO(HInd) 2 (NO 2 ) 2 OH 2 ]AcO•0.5HOAc (1a). Two crystal structures with different solvent environments were determined for the benzimidazole complex: [RuNO(Benzim) 2 (NO 2 ) 2 OH]•H 2 O•EtOH (2) and [RuNO(Benz-Im) 2 (NO 2 ) 2 OH]•2iPrOH (2a). The photolysis processes of NO release of the synthesized compounds in DMSO and phosphate-buffered saline (PBS, pH = 7.4) were investigated using a flow-through system coupled with simultaneous measurement of IR and optical absorption spectra under irradiation with 450 nm light. The quantum yields of NO release in DMSO were determined to be 11.4 ± 0.3% and 6.0 ± 0.2% for HInd and Benz-im complexes, respectively. DFT calculations of the optical absorption spectra and molecular orbitals revealed differences in the HOMO orbitals of the complexes, despite their structural similarity. This finding may explain their quantum yield difference. Complexes before and after photolysis in DMSO/acetonitrile mixture were investigated using the ESI-MS method. Prior to photolysis, the predominant species for both complexes were the initial molecular ions observed either as adducts with Na⁺ or through loss of H⁺, OH⁻, or NO₂⁻. Following photolysis, species resulting from NO loss substituted by water or DMSO were detected. Dark and photoinduced cytotoxicity were studied against human lung carcinoma cells (A549) and non-tumor human lung fibroblasts (MRC-5).Without irradiation, the IC 50 values were 27 ± 3 μM and >75 μM for the A549 line, and 24.9 ± 0.4 μM and >75 μM for the MRC-5 line, for Hind and Benz-im complexes, respectively. Irradiation did not lead to significant changes in cytotoxicity. The lipophilicity tests in PBS/n-octanol were conducted for both complexes to tackle their difference in cell membrane permeability. The logP₃₀₀ values for complexes with HInd and Benz-im were 1.21 and 0.32, respectively. This difference may partially account for the higher permeability of HInd complex resulting in higher cytotoxic effect.</div
p-/n-type conduction in activated carbon NO2 sensors with enhanced low-concentration sensitivity at room temperature
No full textInternational audienceGas sensors play a vital role in monitoring air quality across a wide range of applications, including industry, transport and healthcare. The present study investigates the nitrogen dioxide (NO 2 ) sensing capability of four commercial activated carbons (ACs), including two coal-based (MSC 30 and CW 30) and two bio-based (A supra and PK1-3) ACs at room temperature (25 • C). The ACs exhibited distinct textural properties, with specific surface areas ranging from 916 to 2233 m 2 g -1 . Key sensing parameters including responses, R and R ci (%), response time, recovery time, sensitivity, linearity, repeatability, reversibility and stability were examined under controlled NO 2 exposures. The fabricated sensors were exposed to continuous cycles of varying NO 2 concentrations, from 1 to 10 ppm, in 1 ppm increments, and from 1 to 20 ppm in 5 ppm increments. The sensors exhibited p-or n-type conduction behavior, depending on the AC, confirmed by Mott-Schottky measurements. Reversible sensing was governed by weak physical interactions (physisorption) of NO 2 gas on the sensor surface and charge transport via charge hopping. These findings offer valuable guidance for selecting appropriate materials in the development of high-performance, room temperature AC-based NO 2 sensors, which are an essential component of effective environmental monitoring.</div
Interaction dynamics between epithelial cysts captured by tissue rheology
No full textEpithelial cysts are minimal structures involved in morphogenesis. They are fluid-filled cavities surrounded by an epithelial monolayer. Cysts grow during development and their interactions shape organs. While their growth dynamics as single structures are well characterised, the physical mechanisms underlying their interaction remain poorly understood. Here we design a minimal assay of interacting cyst doublets based on microfabrication, quantitative biology, and theory to show that Madin-Darby Canine Kidney (MDCK) cyst interactions are essentially determined by the rheological properties of their epithelial monolayers. We report two phases of interaction between epithelial cysts: coalescence of cellular monolayers and lumen fusion, with similar speeds of 0.3 μm/h. We modulate the distribution of interaction phenotypes by reducing cell-cell adhesion using E-cadherin knock-out MDCK cells and we report that E-cadherin depletion promotes lumen fusion. Remarkably, the dynamics of coalescence and fusion are conserved between both cell lines. To understand the conserved speeds and the effect of cell-cell adhesion, we model the mechanical behavior of cyst doublets as a complex fluid to predict a speed determined by viscosity, a stretch-dependent monolayer tension, and the adhesion energy between cells. We measure these parameters through rheological experiments using micropipette aspiration and lumen drainage, which span the full range of stretch. A key insight from this analysis is that accounting for the tension dependence on stretch is is essential to capture the different dynamics observed during cyst interaction. Using these rheological measurements, we successfully recapitulate the conserved speed. Altogether, our results open new perspectives to understand tissue dynamics during organogenesis through simple physical arguments
Population coding to improve fault tolerance of neuromorphic networks in regression tasks
No full textInternational audienceSpiking Neural Networks (SNNs) and specialized neuromorphic hardware represent a promising prospect for energy-efficient computation. However, this hardware is susceptible to permanent faults, such as dead or saturated neurons, which can compromise the model's reliability. As semiconductor technologies advance toward ever-smaller feature sizes, process variations and defect rates increase, making fault tolerance a critical requirement. The intrinsic robustness of neural computation, inspired by biological systems, offers an opportunity to develop more sustainable neuromorphic design practices-by enabling the use of partially defective chips both at manufacturing time and during long-term deployment. In this context, we argue that population coding provides an additional layer of fault resilience, as it allows neural models to tolerate hardware-level defects without requiring retraining or architectural modifications. This paper investigates the inherent and passive fault tolerance conferred by population coding as a robustness strategy in regression tasks. We propose a methodology where continuous variables are represented using Gaussian Receptive Field (GRF) population encoding and decoded from the SNN's output using a Maximum Likelihood Estimation (MLE) method designed to mitigate the influence of faulty neurons.We systematically evaluate this approach through fault injection experiments by introducing an increasing number of faults across different network layers. Our results demonstrate that population-coded models may be significantly more resilient to permanent faults than those using a direct single-neuron output. This work validates that population coding provides a powerful architecture for fault-tolerant neuromorphic systems without the overhead of active fault detection and reconfiguration hardware.</p